id
stringlengths 9
16
| submitter
stringlengths 1
64
⌀ | authors
stringlengths 3
62.8k
| title
stringlengths 1
382
| comments
stringlengths 1
1.15k
⌀ | journal-ref
stringlengths 1
557
⌀ | doi
stringlengths 7
153
⌀ | report-no
stringlengths 1
509
⌀ | categories
stringlengths 5
125
| license
stringclasses 9
values | abstract
stringlengths 6
6.09k
| versions
listlengths 1
187
| update_date
timestamp[s]date 2007-05-23 00:00:00
2025-06-27 00:00:00
| authors_parsed
sequencelengths 1
2.83k
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2505.13379 | Gongfan Fang | Gongfan Fang, Xinyin Ma, Xinchao Wang | Thinkless: LLM Learns When to Think | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by/4.0/ | Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning for all queries often results in substantial computational inefficiencies, particularly when many problems admit straightforward solutions. This motivates an open question: Can LLMs learn when to think? To answer this, we propose Thinkless, a learnable framework that empowers an LLM to adaptively select between short-form and long-form reasoning, based on both task complexity and the model's ability. Thinkless is trained under a reinforcement learning paradigm and employs two control tokens, <short> for concise responses and <think> for detailed reasoning. At the core of our method is a Decoupled Group Relative Policy Optimization (DeGRPO) algorithm, which decomposes the learning objective of hybrid reasoning into two components: (1) a control token loss that governs the selection of the reasoning mode, and (2) a response loss that improves the accuracy of the generated answers. This decoupled formulation enables fine-grained control over the contributions of each objective, stabilizing training and effectively preventing collapse observed in vanilla GRPO. Empirically, on several benchmarks such as Minerva Algebra, MATH-500, and GSM8K, Thinkless is able to reduce the usage of long-chain thinking by 50% - 90%, significantly improving the efficiency of Reasoning Language Models. The code is available at https://github.com/VainF/Thinkless | [
{
"version": "v1",
"created": "Mon, 19 May 2025 17:24:16 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:06:49 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Fang",
"Gongfan",
""
],
[
"Ma",
"Xinyin",
""
],
[
"Wang",
"Xinchao",
""
]
] |
2505.14414 | Chengtang Yao | Chengtang Yao, Lidong Yu, Zhidan Liu, Jiaxi Zeng, Yuwei Wu, Yunde Jia | Diving into the Fusion of Monocular Priors for Generalized Stereo Matching | Code: https://github.com/YaoChengTang/Diving-into-the-Fusion-of-Monocular-Priors-for-Generalized-Stereo-Matching | ICCV 2025 | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | The matching formulation makes it naturally hard for the stereo matching to handle ill-posed regions like occlusions and non-Lambertian surfaces. Fusing monocular priors has been proven helpful for ill-posed matching, but the biased monocular prior learned from small stereo datasets constrains the generalization. Recently, stereo matching has progressed by leveraging the unbiased monocular prior from the vision foundation model (VFM) to improve the generalization in ill-posed regions. We dive into the fusion process and observe three main problems limiting the fusion of the VFM monocular prior. The first problem is the misalignment between affine-invariant relative monocular depth and absolute depth of disparity. Besides, when we use the monocular feature in an iterative update structure, the over-confidence in the disparity update leads to local optima results. A direct fusion of a monocular depth map could alleviate the local optima problem, but noisy disparity results computed at the first several iterations will misguide the fusion. In this paper, we propose a binary local ordering map to guide the fusion, which converts the depth map into a binary relative format, unifying the relative and absolute depth representation. The computed local ordering map is also used to re-weight the initial disparity update, resolving the local optima and noisy problem. In addition, we formulate the final direct fusion of monocular depth to the disparity as a registration problem, where a pixel-wise linear regression module can globally and adaptively align them. Our method fully exploits the monocular prior to support stereo matching results effectively and efficiently. We significantly improve the performance from the experiments when generalizing from SceneFlow to Middlebury and Booster datasets while barely reducing the efficiency. | [
{
"version": "v1",
"created": "Tue, 20 May 2025 14:27:45 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Yao",
"Chengtang",
""
],
[
"Yu",
"Lidong",
""
],
[
"Liu",
"Zhidan",
""
],
[
"Zeng",
"Jiaxi",
""
],
[
"Wu",
"Yuwei",
""
],
[
"Jia",
"Yunde",
""
]
] |
2505.14993 | Mihaly Petreczky | Mih\'aly Petreczky and Ziad Alkhoury and Guillaume Merc\`ere | On the equivalence between functionally affine LPV state-space representations and LFT models | null | null | null | null | math.OC | http://creativecommons.org/licenses/by/4.0/ | We propose a transformation algorithm for a class of Linear Parameter-Varying (LPV) systems with functional affine dependence on parameters, where the system matrices depend affinely on nonlinear functions of the scheduling varable, into Linear Fractional Transformation (LFT) systems. The transformation preserves input-output behavior and minimality, and the uncertainity block of the resulting LFT system is linear in the scheduling variables of the LPV system. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 00:43:10 GMT"
},
{
"version": "v2",
"created": "Thu, 22 May 2025 20:05:55 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 02:14:27 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Petreczky",
"Mihály",
""
],
[
"Alkhoury",
"Ziad",
""
],
[
"Mercère",
"Guillaume",
""
]
] |
2505.15858 | Hyeonjoong Cho | HoHyun Sim, Hyeonjoong Cho, Yeonghyeon Go, Zhoulai Fu, Ali Shokri, Binoy Ravindran | Large Language Model-Powered Agent for C to Rust Code Translation | null | null | null | null | cs.PL cs.SE | http://creativecommons.org/licenses/by-nc-sa/4.0/ | The C programming language has been foundational in building system-level software. However, its manual memory management model frequently leads to memory safety issues. In response, a modern system programming language, Rust, has emerged as a memory-safe alternative. Moreover, automating the C-to-Rust translation empowered by the rapid advancements of the generative capabilities of LLMs is gaining growing interest for large volumes of legacy C code. Despite some success, existing LLM-based approaches have constrained the role of LLMs to static prompt-response behavior and have not explored their agentic problem-solving capability. Applying the LLM agentic capability for the C-to-Rust translation introduces distinct challenges, as this task differs from the traditional LLM agent applications, such as math or commonsense QA domains. First, the scarcity of parallel C-to-Rust datasets hinders the retrieval of suitable code translation exemplars for in-context learning. Second, unlike math or commonsense QA, the intermediate steps required for C-to-Rust are not well-defined. Third, it remains unclear how to organize and cascade these intermediate steps to construct a correct translation trajectory. To address these challenges in the C-to-Rust translation, we propose a novel intermediate step, the Virtual Fuzzing-based equivalence Test (VFT), and an agentic planning framework, the LLM-powered Agent for C-to-Rust code translation (LAC2R). The VFT guides LLMs to identify input arguments that induce divergent behaviors between an original C function and its Rust counterpart and to generate informative diagnoses to refine the unsafe Rust code. LAC2R uses the MCTS to systematically organize the LLM-induced intermediate steps for correct translation. We experimentally demonstrated that LAC2R effectively conducts C-to-Rust translation on large-scale, real-world benchmarks. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 01:26:23 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 15:16:53 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Sim",
"HoHyun",
""
],
[
"Cho",
"Hyeonjoong",
""
],
[
"Go",
"Yeonghyeon",
""
],
[
"Fu",
"Zhoulai",
""
],
[
"Shokri",
"Ali",
""
],
[
"Ravindran",
"Binoy",
""
]
] |
2505.16706 | Lia Vas | Lia Vas | The Graded Classification Conjecture holds for graphs with disjoint cycles | This version contains: (1) additional consideration of diagonal preservation, (2) some additional references to related work, (3) some corrections of typos and prerequisites | null | null | null | math.RA math.DS math.OA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The Graded Classification Conjecture (GCC) states that the pointed $K_0^{\operatorname{gr}}$-group is a complete invariant of the Leavitt path algebras of finite graphs when these algebras are considered with their natural grading by $\mathbb Z.$ The conjecture has previously been shown to hold in some special cases. The main result of the paper shows that the GCC holds for a significantly more general class of graphs included in the class of graphs with disjoint cycles. In particular, our result holds for finite graphs with disjoint cycles. We show the main result also for graph $C^*$-algebras. As a consequence, the graded version of the Isomorphism Conjecture holds for the class of graphs we consider.
Besides showing the conjecture for the class of graphs we consider, we realize the Grothendieck $\mathbb Z$-group isomorphism by a specific graded $*$-isomorphism. In particular, we introduce a series of graph operations which preserve the graded $*$-isomorphism class of their algebras. After performing these operations on a graph, we obtain well-behaved ``representative'' graphs, which we call canonical forms. We define an equivalence $\approx$ on graphs such that $E\approx F$ holds when there are isomorphic canonical forms of $E$ and $F$ and we show that the condition $E\approx F$ is equivalent to the existence of an isomorphism $f$ of the Grothendieck $\mathbb Z$-groups of the algebras of $E$ and $F$ in the appropriate category. As $E\approx F$ can be realized by a finite series of specific graph operations, any such isomorphism $f$ can be realized by an explicit graded $*$-algebra isomorphism. Thus, we describe the graded ($*$-)isomorphism classes of the algebras of graphs we consider. Besides the ties to symbolic dynamics and Williams' Problem, such a description is relevant for the active program of classification of graph $C^*$-algebras. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 14:07:50 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:38:27 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Vas",
"Lia",
""
]
] |
2505.16713 | Shogo Nakakita | Shogo Nakakita | Sharp concentration of uniform generalization errors in binary linear classification | 26 pages, 1 figure; minor edits to improve readability | null | null | null | stat.ML cs.LG math.ST stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We examine the concentration of uniform generalization errors around their expectation in binary linear classification problems via an isoperimetric argument. In particular, we establish Poincar\'{e} and log-Sobolev inequalities for the joint distribution of the output labels and the label-weighted input vectors, which we apply to derive concentration bounds. The derived concentration bounds are sharp up to moderate multiplicative constants by those under well-balanced labels. In asymptotic analysis, we also show that almost sure convergence of uniform generalization errors to their expectation occurs in very broad settings, such as proportionally high-dimensional regimes. Using this convergence, we establish uniform laws of large numbers under dimension-free conditions. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 14:14:50 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:57:11 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Nakakita",
"Shogo",
""
]
] |
2505.16946 | Terry Leitch | Sanjana Chalavadi, Andrei Pastor, and Terry Leitch | NY Real Estate Racial Equity Analysis via Applied Machine Learning | updated/replaced stale reference links. Added narrative covering gentrification, racial capitalism, financialization of housing, and segregation. Moved model details to appendices. Added Nivea | null | null | null | cs.CY cs.LG | http://creativecommons.org/licenses/by/4.0/ | This study analyzes tract-level real estate ownership patterns in New York State (NYS) and New York City (NYC) to uncover racial disparities. We use an advanced race/ethnicity imputation model (LSTM+Geo with XGBoost filtering, validated at 89.2% accuracy) to compare the predicted racial composition of property owners to the resident population from census data. We examine both a Full Model (statewide) and a Name-Only LSTM Model (NYC) to assess how incorporating geospatial context affects our predictions and disparity estimates. The results reveal significant inequities: White individuals hold a disproportionate share of properties and property value relative to their population, while Black, Hispanic, and Asian communities are underrepresented as property owners. These disparities are most pronounced in minority-majority neighborhoods, where ownership is predominantly White despite a predominantly non-White population. Corporate ownership (LLCs, trusts, etc.) exacerbates these gaps by reducing owner-occupied opportunities in urban minority communities. We provide a breakdown of ownership vs. population by race for majority-White, -Black, -Hispanic, and -Asian tracts, identify those with extreme ownership disparities, and compare patterns in urban, suburban, and rural contexts. The findings underscore persistent racial inequity in property ownership, reflecting broader historical and socio-economic forces, and highlight the importance of data-driven approaches to address these issues. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 17:32:28 GMT"
},
{
"version": "v2",
"created": "Tue, 03 Jun 2025 03:50:43 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 17:32:06 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Chalavadi",
"Sanjana",
""
],
[
"Pastor",
"Andrei",
""
],
[
"Leitch",
"Terry",
""
]
] |
2505.17186 | Sebastian V\"olkel | Sebastian H. V\"olkel | Bound States of the Schwarzschild Black Hole | 9 pages, 8 figures, 1 table. Matches published version | Phys.Rev.Lett. 134 (2025) 24, 241401 | 10.1103/tbm2-gzv9 | null | gr-qc hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Understanding the physical significance and spectral stability of black hole quasinormal modes is fundamental to high-precision spectroscopy with future gravitational wave detectors. Inspired by Mashhoon's idea of relating quasinormal modes of black holes with their equivalent bound states in an inverted potential, we investigate, for the first time, energy levels and eigenfunctions of the Schwarzschild black hole quantitatively. While quasinormal modes describe the characteristic damped oscillations of a black hole, the bound states of the inverted potential are qualitatively more similar to those of the hydrogen atom. Although the physical interpretation of these states may initially be of more academic interest, it furthers our understanding of open problems related to quasinormal modes in a similar spirit to Maggiore's interpretation of the Schwarzschild quasinormal mode spectrum. One surprising insight from the explicit calculation of bound states is that eigenfunctions corresponding to quasinormal mode overtones become rapidly delocalized and extremely loosely bound. This observation raises immediate questions about the common interpretation of quasinormal modes as excitations of the lightring region. Closely related, as a second application, we also explore the spectral stability of bound states and demonstrate that they can provide complementary insights into the quasinormal mode spectrum. | [
{
"version": "v1",
"created": "Thu, 22 May 2025 18:00:03 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 08:22:36 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Völkel",
"Sebastian H.",
""
]
] |
2505.18389 | Maxime Elkael | Maxime Elkael, Michele Polese, Reshma Prasad, Stefano Maxenti, Tommaso Melodia | ALLSTaR: Automated LLM-Driven Scheduler Generation and Testing for Intent-Based RAN | Under submission to an IEEE journal, copyright may change without notice | null | null | null | cs.NI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The evolution toward open, programmable O-RAN and AI-RAN 6G networks creates unprecedented opportunities for Intent-Based Networking (IBN) to dynamically optimize RAN[...]. However, applying IBN effectively to the RAN scheduler [...] remains a significant challenge. Current approaches predominantly rely on coarse-grained network slicing, lacking the granularity for dynamic adaptation to individual user conditions and traffic patterns. Despite the existence of a vast body of scheduling algorithms [...], their practical utilization is hindered by implementation heterogeneity, insufficient systematic evaluation in production environments, and the complexity of developing high-performance scheduler implementations.[...] To address these limitations, we propose ALLSTaR (Automated LLm-driven Scheduler generation and Testing for intent-based RAN), a novel framework leveraging LLMs for automated, intent-driven scheduler design, implementation, and evaluation. ALLSTaR interprets NL intents, automatically generates functional scheduler code from the research literature using OCR and LLMs, and intelligently matches operator intents to the most suitable scheduler(s). Our implementation deploys these schedulers as O-RAN dApps, enabling on-the-fly deployment and testing on a production-grade, 5G-compliant testbed. This approach has enabled the largest-scale OTA experimental comparison of 18 scheduling algorithms automatically synthesized from the academic literature. The resulting performance profiles serve as the input for our Intent-Based Scheduling (IBS) framework, which dynamically selects and deploys appropriate schedulers that optimally satisfy operator intents. We validate our approach through multiple use cases unattainable with current slicing-based optimization techniques, demonstrating fine-grained control based on buffer status, physical layer conditions, and heterogeneous traffic types | [
{
"version": "v1",
"created": "Fri, 23 May 2025 21:33:16 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 14:13:53 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 04:31:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Elkael",
"Maxime",
""
],
[
"Polese",
"Michele",
""
],
[
"Prasad",
"Reshma",
""
],
[
"Maxenti",
"Stefano",
""
],
[
"Melodia",
"Tommaso",
""
]
] |
2505.19129 | Yonghui Xia | Yuchao He, Mengda Wu, Yonghui Xia, Meirong Zhang | A novel and application-oriented inverse nodal problem for Sturm-Liouville operators | null | null | null | null | math.CA | http://creativecommons.org/licenses/by/4.0/ | This paper develops a methodological framework for addressing a novel and application-oriented inverse nodal problem in Sturm-Liouville operators, having significant applications in seismic wave analysis and submarine underwater radar (sonar) detection. By utilizing a given finite set of nodal data, we propose an optimization framework to find the potential $\hat q$ that is most closely approximating a predefined target potential $q_0$. The inverse nodal optimization problem is reformulated as a solvability problem for a class of nonlinear Schr\"odinger equations, enabling systematic investigation of the inverse nodal problem. {As an example, when the constant target potential $q_0$ is considered, we find that the Schr\"odinger equations are completely integrable and conclude that the potential $\hat q$ is `periodic' in a certain sense. Furthermore, the reconstruction of $\hat q$ is reduced to solving a system of three featured parameters, thereby establishing an explicit quantitative relationship between $\|\hat q\|_{Lp}$ and $T_*$. Of importance, we prove the uniqueness of the potential $\hat q$ when $p>3/2$. These new findings represent a substantial advancement in this field of study. Our methodology also bridges theoretical rigor with practical applicability, addressing scenarios where only partial nodal information is available. | [
{
"version": "v1",
"created": "Sun, 25 May 2025 12:53:54 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:51:57 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"He",
"Yuchao",
""
],
[
"Wu",
"Mengda",
""
],
[
"Xia",
"Yonghui",
""
],
[
"Zhang",
"Meirong",
""
]
] |
2505.19197 | Jihoon Kwon | Chanyeol Choi, Alejandro Lopez-Lira, Yongjae Lee, Jihoon Kwon, Minjae Kim, Juneha Hwang, Minsoo Ha, Chaewoon Kim, Jaeseon Ha, Suyeol Yun, Jin Kim | Structuring the Unstructured: A Multi-Agent System for Extracting and Querying Financial KPIs and Guidance | 7 pages, FinIR'25 | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Extracting structured and quantitative insights from unstructured financial filings is essential in investment research, yet remains time-consuming and resource-intensive. Conventional approaches in practice rely heavily on labor-intensive manual processes, limiting scalability and delaying the research workflow. In this paper, we propose an efficient and scalable method for accurately extracting quantitative insights from unstructured financial documents, leveraging a multi-agent system composed of large language models. Our proposed multi-agent system consists of two specialized agents: the \emph{Extraction Agent} and the \emph{Text-to-SQL Agent}. The \textit{Extraction Agent} automatically identifies key performance indicators from unstructured financial text, standardizes their formats, and verifies their accuracy. On the other hand, the \textit{Text-to-SQL Agent} generates executable SQL statements from natural language queries, allowing users to access structured data accurately without requiring familiarity with the database schema. Through experiments, we demonstrate that our proposed system effectively transforms unstructured text into structured data accurately and enables precise retrieval of key information. First, we demonstrate that our system achieves approximately 95\% accuracy in transforming financial filings into structured data, matching the performance level typically attained by human annotators. Second, in a human evaluation of the retrieval task -- where natural language queries are used to search information from structured data -- 91\% of the responses were rated as correct by human evaluators. In both evaluations, our system generalizes well across financial document types, consistently delivering reliable performance. | [
{
"version": "v1",
"created": "Sun, 25 May 2025 15:45:46 GMT"
},
{
"version": "v2",
"created": "Tue, 27 May 2025 13:32:03 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 04:56:31 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Choi",
"Chanyeol",
""
],
[
"Lopez-Lira",
"Alejandro",
""
],
[
"Lee",
"Yongjae",
""
],
[
"Kwon",
"Jihoon",
""
],
[
"Kim",
"Minjae",
""
],
[
"Hwang",
"Juneha",
""
],
[
"Ha",
"Minsoo",
""
],
[
"Kim",
"Chaewoon",
""
],
[
"Ha",
"Jaeseon",
""
],
[
"Yun",
"Suyeol",
""
],
[
"Kim",
"Jin",
""
]
] |
2505.20574 | Can Polat | Can Polat, Mehmet Tuncel, Mustafa Kurban, Erchin Serpedin, Hasan Kurban | xChemAgents: Agentic AI for Explainable Quantum Chemistry | Accepted Paper at ICML 2025 Workshop on MAS | null | null | null | cs.MA physics.chem-ph physics.comp-ph | http://creativecommons.org/licenses/by/4.0/ | Recent progress in multimodal graph neural networks has demonstrated that augmenting atomic XYZ geometries with textual chemical descriptors can enhance predictive accuracy across a range of electronic and thermodynamic properties. However, naively appending large sets of heterogeneous descriptors often degrades performance on tasks sensitive to molecular shape or symmetry, and undermines interpretability. xChemAgents proposes a cooperative agent framework that injects physics-aware reasoning into multimodal property prediction. xChemAgents comprises two language-model-based agents: a Selector, which adaptively identifies a sparse, weighted subset of descriptors relevant to each target, and provides a natural language rationale; and a Validator, which enforces physical constraints such as unit consistency and scaling laws through iterative dialogue. On standard benchmark datasets, xChemAgents achieves up to a 22% reduction in mean absolute error over the state-of-the-art baselines, while producing faithful, human-interpretable explanations. Experiment results highlight the potential of cooperative, self-verifying agents to enhance both accuracy and transparency in foundation-model-driven materials science. The implementation and accompanying dataset are available at https://github.com/KurbanIntelligenceLab/xChemAgents. | [
{
"version": "v1",
"created": "Mon, 26 May 2025 23:22:41 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:07:02 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Polat",
"Can",
""
],
[
"Tuncel",
"Mehmet",
""
],
[
"Kurban",
"Mustafa",
""
],
[
"Serpedin",
"Erchin",
""
],
[
"Kurban",
"Hasan",
""
]
] |
2505.21348 | Yang Liu | Shunrui Li and Yang Liu | Quantum harmonic oscillator, index theorem and anomaly | 1 figure | null | null | null | quant-ph hep-th | http://creativecommons.org/licenses/by/4.0/ | We report a bosonic anomaly emerging in the quantum harmonic oscillator, whose partition function is rigorously identified as the Chern character via the Grothendieck-Riemann-Roch theorem, establishing a new connection among statistical mechanics, anomaly, Atiyah-Singer index theorem and Gromov-Witten theory. We investigate how its internal energy relates to the Atiyah-Singer index theorem, showing that the partition function can be interpreted as the Chern character of "physical sheaf" over Eucildean spacetime by using Grothentic-Riemann-Roch theorem. This correspondence reveals the internal energy of oscillator as a concrete non-SUSY manifestation of the index theorem. Moreover, we show that this connection naturally leads to the emergence of a quantum anomaly. Furthermore, we arrive at Gromov-Witten theory through a more direct and physically intuitive approach. As a result, the internal energy of the quantum harmonic oscillator serves as a bridge linking two key concepts in physics -- statistical mechanics and anomalies -- with three fundamental mathematical frameworks: the Atiyah-Singer index theorem, the Grothendieck-Riemann-Roch theorem, and Gromov-Witten theory. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 15:43:35 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:44:33 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Li",
"Shunrui",
""
],
[
"Liu",
"Yang",
""
]
] |
2505.21657 | Zeinab Dehghani | Zeinab Dehghani, Mohammed Naveed Akram, Koorosh Aslansefat and Adil Khan | Explainability of Large Language Models using SMILE: Statistical Model-agnostic Interpretability with Local Explanations | The submission contains incorrect references that require substantial revision | null | null | null | cs.CL cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large language models like GPT, LLAMA, and Claude have become incredibly powerful at generating text, but they are still black boxes, so it is hard to understand how they decide what to say. That lack of transparency can be problematic, especially in fields where trust and accountability matter. To help with this, we introduce SMILE, a new method that explains how these models respond to different parts of a prompt. SMILE is model-agnostic and works by slightly changing the input, measuring how the output changes, and then highlighting which words had the most impact. Create simple visual heat maps showing which parts of a prompt matter the most. We tested SMILE on several leading LLMs and used metrics such as accuracy, consistency, stability, and fidelity to show that it gives clear and reliable explanations. By making these models easier to understand, SMILE brings us one step closer to making AI more transparent and trustworthy. | [
{
"version": "v1",
"created": "Tue, 27 May 2025 18:32:38 GMT"
},
{
"version": "v2",
"created": "Fri, 13 Jun 2025 16:43:15 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 16:16:59 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Dehghani",
"Zeinab",
""
],
[
"Akram",
"Mohammed Naveed",
""
],
[
"Aslansefat",
"Koorosh",
""
],
[
"Khan",
"Adil",
""
]
] |
2505.22246 | Nedko Savov | Nedko Savov, Naser Kazemi, Deheng Zhang, Danda Pani Paudel, Xi Wang, Luc Van Gool | StateSpaceDiffuser: Bringing Long Context to Diffusion World Models | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | World models have recently become promising tools for predicting realistic visuals based on actions in complex environments. However, their reliance on only a few recent observations leads them to lose track of the long-term context. Consequently, in just a few steps the generated scenes drift from what was previously observed, undermining the temporal coherence of the sequence. This limitation of the state-of-the-art world models, most of which rely on diffusion, comes from their lack of a lasting environment state. To address this problem, we introduce StateSpaceDiffuser, where a diffusion model is enabled to perform long-context tasks by integrating features from a state-space model, representing the entire interaction history. This design restores long-term memory while preserving the high-fidelity synthesis of diffusion models. To rigorously measure temporal consistency, we develop an evaluation protocol that probes a model's ability to reinstantiate seen content in extended rollouts. Comprehensive experiments show that StateSpaceDiffuser significantly outperforms a strong diffusion-only baseline, maintaining a coherent visual context for an order of magnitude more steps. It delivers consistent views in both a 2D maze navigation and a complex 3D environment. These results establish that bringing state-space representations into diffusion models is highly effective in demonstrating both visual details and long-term memory. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 11:27:54 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:10:36 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Savov",
"Nedko",
""
],
[
"Kazemi",
"Naser",
""
],
[
"Zhang",
"Deheng",
""
],
[
"Paudel",
"Danda Pani",
""
],
[
"Wang",
"Xi",
""
],
[
"Van Gool",
"Luc",
""
]
] |
2505.22874 | Navindra Singh | Navindra D. Singh, James Leung, Ji Feng, Alma K. Gonz\'alez-Alcalde, Arial Tolentino, David Tuft, Juchen Guo, and Luat T. Vuong | Spectrum Selective Interfaces and Materials towards Non-photothermal Saltwater Evaporation: Demonstration with a White Ceramic Wick | 15 pages, 5 figures, submitted to ACS ES&T, 68 references | null | null | null | cond-mat.mtrl-sci physics.optics | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Most solar desalination efforts are photothermal: they evaporate water with ``black'' materials that absorb as much sunlight as possible. Such ``brine-boiling'' methods are severely limited by the high thermal mass of water, i.e., its capacity to store and release heat. Here, we study the light-enhanced evaporation by a hard, white, aluminum nitride wick, which reveals a route to selectively target salt-water bonds instead of bulk heating. Evaporation rates dramatically increase with short-wavelength illumination. Violet-light illumination achieves 4-10x higher evaporation enhancement compared to orange and IR light. Our results identify a light-driven, spectrum-selective path to non-photothermal saltwater evaporation and opportunities to employ ceramic wicks for salt harvesting. Such low-cost, low-energy desalination systems would reduce the heat island effects of traditional solar technologies and contribute to new cooling technologies where drought is also a concern. | [
{
"version": "v1",
"created": "Wed, 28 May 2025 21:16:44 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:02:55 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Singh",
"Navindra D.",
""
],
[
"Leung",
"James",
""
],
[
"Feng",
"Ji",
""
],
[
"González-Alcalde",
"Alma K.",
""
],
[
"Tolentino",
"Arial",
""
],
[
"Tuft",
"David",
""
],
[
"Guo",
"Juchen",
""
],
[
"Vuong",
"Luat T.",
""
]
] |
2505.23062 | Lingkai Kong | Lingkai Kong, Haichuan Wang, Tonghan Wang, Guojun Xiong, Milind Tambe | Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Incorporating pre-collected offline data from a source environment can significantly improve the sample efficiency of reinforcement learning (RL), but this benefit is often challenged by discrepancies between the transition dynamics of the source and target environments. Existing methods typically address this issue by penalizing or filtering out source transitions in high dynamics-gap regions. However, their estimation of the dynamics gap often relies on KL divergence or mutual information, which can be ill-defined when the source and target dynamics have disjoint support. To overcome these limitations, we propose CompFlow, a method grounded in the theoretical connection between flow matching and optimal transport. Specifically, we model the target dynamics as a conditional flow built upon the output distribution of the source-domain flow, rather than learning it directly from a Gaussian prior. This composite structure offers two key advantages: (1) improved generalization for learning target dynamics, and (2) a principled estimation of the dynamics gap via the Wasserstein distance between source and target transitions. Leveraging our principled estimation of the dynamics gap, we further introduce an optimistic active data collection strategy that prioritizes exploration in regions of high dynamics gap, and theoretically prove that it reduces the performance disparity with the optimal policy. Empirically, CompFlow outperforms strong baselines across several RL benchmarks with shifted dynamics. | [
{
"version": "v1",
"created": "Thu, 29 May 2025 04:09:19 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:09:46 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kong",
"Lingkai",
""
],
[
"Wang",
"Haichuan",
""
],
[
"Wang",
"Tonghan",
""
],
[
"Xiong",
"Guojun",
""
],
[
"Tambe",
"Milind",
""
]
] |
2505.24466 | Yingjia Xu | Yingjia Xu, Jinlin Wu, Zhen Chen, Daming Gao, Yang Yang, Zhen Lei, Min Cao | SA-Person: Text-Based Person Retrieval with Scene-aware Re-ranking | 22 pages, 7 figures. Under review | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Text-based person retrieval aims to identify a target individual from a gallery of images based on a natural language description. It presents a significant challenge due to the complexity of real-world scenes and the ambiguity of appearance-related descriptions. Existing methods primarily emphasize appearance-based cross-modal retrieval, often neglecting the contextual information embedded within the scene, which can offer valuable complementary insights for retrieval. To address this, we introduce SCENEPERSON-13W, a large-scale dataset featuring over 100,000 scenes with rich annotations covering both pedestrian appearance and environmental cues. Based on this, we propose SA-Person, a two-stage retrieval framework. In the first stage, it performs discriminative appearance grounding by aligning textual cues with pedestrian-specific regions. In the second stage, it introduces SceneRanker, a training-free, scene-aware re-ranking method leveraging multimodal large language models to jointly reason over pedestrian appearance and the global scene context. Experiments on SCENEPERSON-13W validate the effectiveness of our framework in challenging scene-level retrieval scenarios. The code and dataset will be made publicly available. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 11:10:28 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:46:10 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Xu",
"Yingjia",
""
],
[
"Wu",
"Jinlin",
""
],
[
"Chen",
"Zhen",
""
],
[
"Gao",
"Daming",
""
],
[
"Yang",
"Yang",
""
],
[
"Lei",
"Zhen",
""
],
[
"Cao",
"Min",
""
]
] |
2505.24494 | Xia Zhao | Xia Zhao and Peibiao Zhao | The Minkowski problem for the $k$-torsional rigidity | null | null | null | null | math.DG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | P. Salani [Adv. Math., 229 (2012)] introduced the $k$-torsional rigidity associated with a $k$-Hessian equation and obtained the Brunn-Minkowski inequalities $w.r.t.$ the torsional rigidity in $\mathbb{R}^3$. Following this work, we first construct, in the present paper, a Hadamard variational formula for the $k$-torsional rigidity with $1\leq k\leq n-1$, then we can deduce a $k$-torsional measure from the Hadamard variational formula. Based on the $k$-torsional measure, we propose the Minkowski problem for the $k$-torsional rigidity and confirm the existence of its smooth non-even solutions by the method of a curvature flow. Specially, a new proof method for the uniform lower bound estimation in the $C^0$ estimation for the solution to the curvature flow is presented with the help of invariant functional $\Phi(\Omega_t)$. | [
{
"version": "v1",
"created": "Fri, 30 May 2025 11:46:27 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 11:58:04 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Zhao",
"Xia",
""
],
[
"Zhao",
"Peibiao",
""
]
] |
2506.01244 | Henrik Rosenberger | Henrik Rosenberger, Benjamin Sanderse, Giovanni Stabile | Exact operator inference with minimal data | null | null | null | null | math.NA cs.NA | http://creativecommons.org/licenses/by-nc-nd/4.0/ | This work introduces a novel method to generate snapshot data for operator inference that guarantees the exact reconstruction of intrusive projection-based reduced-order models (ROMs). To ensure exact reconstruction, the operator inference least squares matrix must have full rank, without regularization. Existing works have achieved this full rank using heuristic strategies to generate snapshot data and a-posteriori checks on full rank, but without a guarantee of success. Our novel snapshot data generation method provides this guarantee thanks to two key ingredients: first we identify ROM states that induce full rank, then we generate snapshots corresponding to exactly these states by simulating multiple trajectories for only a single time step. This way, the number of required snapshots is minimal and orders of magnitude lower than typically reported with existing methods. The method avoids non-Markovian terms and does not require re-projection. Since the number of snapshots is minimal, the least squares problem simplifies to a linear system that is numerically more stable. In addition, because the inferred operators are exact, properties of the intrusive ROM operators such as symmetry or skew-symmetry are preserved. Numerical results for differential equations involving 2nd, 3rd and 8th order polynomials demonstrate that the novel snapshot data generation method leads to exact reconstruction of the intrusive reduced order models. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 01:44:33 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:44:50 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Rosenberger",
"Henrik",
""
],
[
"Sanderse",
"Benjamin",
""
],
[
"Stabile",
"Giovanni",
""
]
] |
2506.01495 | Ping Wu | Ping Wu, Guobin Shen, Dongcheng Zhao, Yuwei Wang, Yiting Dong, Yu Shi, Enmeng Lu, Feifei Zhao, Yi Zeng | CVC: A Large-Scale Chinese Value Rule Corpus for Value Alignment of Large Language Models | null | null | null | null | cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Ensuring that Large Language Models (LLMs) align with mainstream human values and ethical norms is crucial for the safe and sustainable development of AI. Current value evaluation and alignment are constrained by Western cultural bias and incomplete domestic frameworks reliant on non-native rules; furthermore, the lack of scalable, rule-driven scenario generation methods makes evaluations costly and inadequate across diverse cultural contexts. To address these challenges, we propose a hierarchical value framework grounded in core Chinese values, encompassing three main dimensions, 12 core values, and 50 derived values. Based on this framework, we construct a large-scale Chinese Values Corpus (CVC) containing over 250,000 value rules enhanced and expanded through human annotation. Experimental results show that CVC-guided scenarios outperform direct generation ones in value boundaries and content diversity. In the evaluation across six sensitive themes (e.g., surrogacy, suicide), seven mainstream LLMs preferred CVC-generated options in over 70.5% of cases, while five Chinese human annotators showed an 87.5% alignment with CVC, confirming its universality, cultural relevance, and strong alignment with Chinese values. Additionally, we construct 400,000 rule-based moral dilemma scenarios that objectively capture nuanced distinctions in conflicting value prioritization across 17 LLMs. Our work establishes a culturally-adaptive benchmarking framework for comprehensive value evaluation and alignment, representing Chinese characteristics. All data are available at https://huggingface.co/datasets/Beijing-AISI/CVC, and the code is available at https://github.com/Beijing-AISI/CVC. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 09:56:59 GMT"
},
{
"version": "v2",
"created": "Sat, 07 Jun 2025 07:06:31 GMT"
},
{
"version": "v3",
"created": "Thu, 19 Jun 2025 13:47:55 GMT"
},
{
"version": "v4",
"created": "Thu, 26 Jun 2025 11:34:33 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wu",
"Ping",
""
],
[
"Shen",
"Guobin",
""
],
[
"Zhao",
"Dongcheng",
""
],
[
"Wang",
"Yuwei",
""
],
[
"Dong",
"Yiting",
""
],
[
"Shi",
"Yu",
""
],
[
"Lu",
"Enmeng",
""
],
[
"Zhao",
"Feifei",
""
],
[
"Zeng",
"Yi",
""
]
] |
2506.01570 | Ding Wen | Deng-Shan Wang and Ding Wen | Long-time asymptotics of the defocusing mKdV equation with step initial data | 41 pages, 19 figures | null | null | null | math.AP math-ph math.MP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This work investigates the long-time asymptotics of solution to defocusing modified Korteweg-de Vries equation with a class of step initial data. A rigorous asymptotic analysis is conducted on the associated Riemann-Hilbert problem by applying Deift-Zhou nonlinear steepest descent method. In this process, the construction of odd-symmetry g-function is generalized and the method of genus reduction on the Riemann-theta function is proposed via conformal transformation and symmetries. It is revealed that for sufficiently large time, the solution manifests a tripartite spatiotemporal structure, i.e., in the left plane-wave region, the solution decays to a modulated plane wave with oscillatory correction; in the central dispersive shock wave region, the solution is governed by a modulated elliptic periodic wave; in the right plane wave region, the solution converges exponentially to a constant. The results from the long-time asymptotic analysis have been shown to match remarkably well with that obtained by direct numerical simulations. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 11:54:12 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:23:01 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Deng-Shan",
""
],
[
"Wen",
"Ding",
""
]
] |
2506.01923 | Amin Karimi Monsefi | Amin Karimi Monsefi, Mridul Khurana, Rajiv Ramnath, Anuj Karpatne, Wei-Lun Chao, Cheng Zhang | TaxaDiffusion: Progressively Trained Diffusion Model for Fine-Grained Species Generation | Accepted to ICCV 2025 | null | null | null | cs.CV cs.AI cs.LG | http://creativecommons.org/publicdomain/zero/1.0/ | We propose TaxaDiffusion, a taxonomy-informed training framework for diffusion models to generate fine-grained animal images with high morphological and identity accuracy. Unlike standard approaches that treat each species as an independent category, TaxaDiffusion incorporates domain knowledge that many species exhibit strong visual similarities, with distinctions often residing in subtle variations of shape, pattern, and color. To exploit these relationships, TaxaDiffusion progressively trains conditioned diffusion models across different taxonomic levels -- starting from broad classifications such as Class and Order, refining through Family and Genus, and ultimately distinguishing at the Species level. This hierarchical learning strategy first captures coarse-grained morphological traits shared by species with common ancestors, facilitating knowledge transfer before refining fine-grained differences for species-level distinction. As a result, TaxaDiffusion enables accurate generation even with limited training samples per species. Extensive experiments on three fine-grained animal datasets demonstrate that outperforms existing approaches, achieving superior fidelity in fine-grained animal image generation. Project page: https://amink8.github.io/TaxaDiffusion/ | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 17:43:55 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:02:25 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Monsefi",
"Amin Karimi",
""
],
[
"Khurana",
"Mridul",
""
],
[
"Ramnath",
"Rajiv",
""
],
[
"Karpatne",
"Anuj",
""
],
[
"Chao",
"Wei-Lun",
""
],
[
"Zhang",
"Cheng",
""
]
] |
2506.02274 | Leo Hahn | Leo Hahn | A General Approach to the Shape Transition of Run-and-Tumble Particles: The 1D PDMP Framework for Invariant Measure Regularity | null | null | null | null | cond-mat.stat-mech math.PR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Run-and-tumble particles (RTPs) have emerged as a paradigmatic example for studying nonequilibrium phenomena in statistical mechanics. The invariant measure of a wide class of RTPs subjected to a potential possesses a density that is continuous at high tumble rates but exhibits divergences at low ones. This key feature, known as shape transition, constitutes a qualitative indicator of the relative closeness (continuous density) or strong deviation (diverging density) from the equilibrium setting. Furthermore, the points at which the density diverges correspond to the configurations where the system spends most of its time in the low tumble rate regime. Building on and extending existing results concerning the regularity of the invariant measure of one-dimensional piecewise-deterministic Markov processes (PDMPs), we show how to characterize the shape transition even in situations where the invariant measure cannot be computed explicitly. Our analysis confirms shape transition as a robust, general feature of RTPs subjected to a potential. We also refine the regularity theory for the invariant measure of one-dimensional PDMPs. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 21:32:57 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 15:06:17 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Hahn",
"Leo",
""
]
] |
2506.02280 | Kedir Yassin H | Kedir Yassin Hussen, Walelign Tewabe Sewunetie, Abinew Ali Ayele, Sukairaj Hafiz Imam, Shamsuddeen Hassan Muhammad, Seid Muhie Yimam | The State of Large Language Models for African Languages: Progress and Challenges | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) are transforming Natural Language Processing (NLP), but their benefits are largely absent for Africa's 2,000 low-resource languages. This paper comparatively analyzes African language coverage across six LLMs, eight Small Language Models (SLMs), and six Specialized SLMs (SSLMs). The evaluation covers language coverage, training sets, technical limitations, script problems, and language modelling roadmaps. The work identifies 42 supported African languages and 23 available public data sets, and it shows a big gap where four languages (Amharic, Swahili, Afrikaans, and Malagasy) are always treated while there is over 98\% of unsupported African languages. Moreover, the review shows that just Latin, Arabic, and Ge'ez scripts are identified while 20 active scripts are neglected. Some of the primary challenges are lack of data, tokenization biases, computational costs being very high, and evaluation issues. These issues demand language standardization, corpus development by the community, and effective adaptation methods for African languages. | [
{
"version": "v1",
"created": "Mon, 02 Jun 2025 21:39:40 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 16:31:32 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 01:01:18 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Hussen",
"Kedir Yassin",
""
],
[
"Sewunetie",
"Walelign Tewabe",
""
],
[
"Ayele",
"Abinew Ali",
""
],
[
"Imam",
"Sukairaj Hafiz",
""
],
[
"Muhammad",
"Shamsuddeen Hassan",
""
],
[
"Yimam",
"Seid Muhie",
""
]
] |
2506.02751 | Chuanyu Fu | Chuanyu Fu, Yuqi Zhang, Kunbin Yao, Guanying Chen, Yuan Xiong, Chuan Huang, Shuguang Cui, Xiaochun Cao | RobustSplat: Decoupling Densification and Dynamics for Transient-Free 3DGS | ICCV 2025. Project page: https://fcyycf.github.io/RobustSplat/ | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | 3D Gaussian Splatting (3DGS) has gained significant attention for its real-time, photo-realistic rendering in novel-view synthesis and 3D modeling. However, existing methods struggle with accurately modeling scenes affected by transient objects, leading to artifacts in the rendered images. We identify that the Gaussian densification process, while enhancing scene detail capture, unintentionally contributes to these artifacts by growing additional Gaussians that model transient disturbances. To address this, we propose RobustSplat, a robust solution based on two critical designs. First, we introduce a delayed Gaussian growth strategy that prioritizes optimizing static scene structure before allowing Gaussian splitting/cloning, mitigating overfitting to transient objects in early optimization. Second, we design a scale-cascaded mask bootstrapping approach that first leverages lower-resolution feature similarity supervision for reliable initial transient mask estimation, taking advantage of its stronger semantic consistency and robustness to noise, and then progresses to high-resolution supervision to achieve more precise mask prediction. Extensive experiments on multiple challenging datasets show that our method outperforms existing methods, clearly demonstrating the robustness and effectiveness of our method. Our project page is https://fcyycf.github.io/RobustSplat/. | [
{
"version": "v1",
"created": "Tue, 03 Jun 2025 11:13:48 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:46:51 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Fu",
"Chuanyu",
""
],
[
"Zhang",
"Yuqi",
""
],
[
"Yao",
"Kunbin",
""
],
[
"Chen",
"Guanying",
""
],
[
"Xiong",
"Yuan",
""
],
[
"Huang",
"Chuan",
""
],
[
"Cui",
"Shuguang",
""
],
[
"Cao",
"Xiaochun",
""
]
] |
2506.04188 | Nicola Guglielmi | Nicola Guglielmi and Ernst Hairer | A fast and memoryless numerical method for solving fractional differential equations | null | null | null | null | math.NA cs.NA math.DS | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The numerical solution of implicit and stiff differential equations by implicit numerical integrators has been largely investigated and there exist many excellent efficient codes available in the scientific community, as Radau5 (based on a Runge-Kutta collocation method at Radau points) and Dassl, based on backward differentiation formulas, among the others. When solving fractional ordinary differential equations (ODEs), the derivative operator is replaced by a non-local one and the fractional ODE is reformulated as a Volterra integral equation, to which these codes cannot be directly applied.
This article is a follow-up of the article by the authors (Guglielmi and Hairer, SISC, 2025) for differential equations with distributed delays. The main idea is to approximate the fractional kernel $t^{\alpha -1}/ \Gamma (\alpha )$ ($\alpha >0$) by a sum of exponential functions or by a sum of exponential functions multiplied by a monomial, and then to transform the fractional integral (of convolution type) into a set of ordinary differential equations. The augmented system is typically stiff and thus requires the use of an implicit method. It can have a very large dimension and requires a special treatment of the arising linear systems.
The present work presents an algorithm for the construction of an approximation of the fractional kernel by a sum of exponential functions, and it shows how the arising linear systems in a stiff time integrator can be solved efficiently. It is explained how the code Radau5 can be used for solving fractional differential equations. Numerical experiments illustrate the accuracy and the efficiency of the proposed method. Driver examples are publicly available from the homepages of the authors. | [
{
"version": "v1",
"created": "Wed, 04 Jun 2025 17:36:01 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:36:34 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Guglielmi",
"Nicola",
""
],
[
"Hairer",
"Ernst",
""
]
] |
2506.04202 | Yanting Wang | Yanting Wang, Wei Zou, Runpeng Geng, Jinyuan Jia | TracLLM: A Generic Framework for Attributing Long Context LLMs | To appear in USENIX Security Symposium 2025. The code and data are at: https://github.com/Wang-Yanting/TracLLM | null | null | null | cs.CR cs.AI cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Long context large language models (LLMs) are deployed in many real-world applications such as RAG, agent, and broad LLM-integrated applications. Given an instruction and a long context (e.g., documents, PDF files, webpages), a long context LLM can generate an output grounded in the provided context, aiming to provide more accurate, up-to-date, and verifiable outputs while reducing hallucinations and unsupported claims. This raises a research question: how to pinpoint the texts (e.g., sentences, passages, or paragraphs) in the context that contribute most to or are responsible for the generated output by an LLM? This process, which we call context traceback, has various real-world applications, such as 1) debugging LLM-based systems, 2) conducting post-attack forensic analysis for attacks (e.g., prompt injection attack, knowledge corruption attacks) to an LLM, and 3) highlighting knowledge sources to enhance the trust of users towards outputs generated by LLMs. When applied to context traceback for long context LLMs, existing feature attribution methods such as Shapley have sub-optimal performance and/or incur a large computational cost. In this work, we develop TracLLM, the first generic context traceback framework tailored to long context LLMs. Our framework can improve the effectiveness and efficiency of existing feature attribution methods. To improve the efficiency, we develop an informed search based algorithm in TracLLM. We also develop contribution score ensemble/denoising techniques to improve the accuracy of TracLLM. Our evaluation results show TracLLM can effectively identify texts in a long context that lead to the output of an LLM. Our code and data are at: https://github.com/Wang-Yanting/TracLLM. | [
{
"version": "v1",
"created": "Wed, 04 Jun 2025 17:48:16 GMT"
},
{
"version": "v2",
"created": "Fri, 06 Jun 2025 02:32:31 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 16:09:36 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Yanting",
""
],
[
"Zou",
"Wei",
""
],
[
"Geng",
"Runpeng",
""
],
[
"Jia",
"Jinyuan",
""
]
] |
2506.04796 | Itai Epstein | Tomer Eini, N. M. R. Peres, Yarden Mazor, and Itai Epstein | Strongly Coupled Exciton--Hyperbolic-phonon-polariton Hybridized States in hBN-encapsulated Biased Bilayer Graphene | arXiv admin note: substantial text overlap with arXiv:2412.03139 | null | null | null | physics.optics cond-mat.mes-hall | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Excitons in biased bilayer graphene are electrically tunable optical excitations residing in the mid-infrared (MIR) spectral range, where intrinsic optical transitions are typically scarce. Such a tunable material system with an excitonic response offer a rare platform for exploring light-matter interactions and optical hybridization of quasiparticles residing in the long wavelength spectrum. In this work, we demonstrate that when the bilayer is encapsulated in hexagonal-boron-nitride (hBN)-a material supporting optical phonons and hyperbolic-phonon-polaritons (HPhPs) in the MIR-the excitons can be tuned into resonance with the HPhP modes. We find that the overlap in energy and momentum of the two MIR quasiparticles facilitate the formation of multiple strongly coupled hybridized exciton-HPhP states. Using an electromagnetic transmission line model, we derive the dispersion relations of the hybridized states and show that they are highly affected and can be manipulated by the symmetry of the system, determining the hybridization selection rules. Our results establish a general tunable MIR platform for engineering strongly coupled quasiparticle states in biased graphene systems, opening new directions for studying and controlling light-matter interactions in the long-wavelength regime. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 09:23:02 GMT"
},
{
"version": "v2",
"created": "Fri, 06 Jun 2025 04:20:42 GMT"
},
{
"version": "v3",
"created": "Sun, 22 Jun 2025 09:21:25 GMT"
},
{
"version": "v4",
"created": "Tue, 24 Jun 2025 18:25:56 GMT"
},
{
"version": "v5",
"created": "Thu, 26 Jun 2025 07:09:52 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Eini",
"Tomer",
""
],
[
"Peres",
"N. M. R.",
""
],
[
"Mazor",
"Yarden",
""
],
[
"Epstein",
"Itai",
""
]
] |
2506.04846 | Piotr Lebiedowicz | Piotr Lebiedowicz, Otto Nachtmann, Antoni Szczurek | Central exclusive production of $\eta$ and $\eta'$ mesons in diffractive proton-proton collisions at the LHC within the tensor-pomeron approach | 17 pages, 8 figures | null | null | null | hep-ph hep-ex | http://creativecommons.org/licenses/by/4.0/ | We present a study of the central exclusive production (CEP) of $\eta$ and $\eta'(958)$ mesons in diffractive proton-proton collisions at high energies. The amplitudes, including pomeron and $f_{2 \mathbb{R}}$ reggeon exchanges, are calculated within the tensor-pomeron model. Absorption effects are also taken into account at the amplitude level. We fit some undetermined model parameters (coupling constants and cutoff parameters in form factors) to the WA102 experimental data and then make predictions for the LHC energy $\sqrt{s} = 13$ TeV. Both, total cross sections and several differential distributions are presented. For $pp \to pp \eta$, we find an upper limit for the total cross section of 2.5 $\mu$b for pseudorapidity of the $\eta$ meson $|\eta_{M}| < 1$ and 5.6 $\mu$b for $2 < \eta_{M} < 5$. For $pp \to pp \eta'$, we predict the cross section to be in the range of 0.3-0.7 $\mu$b for pseudorapidity of the $\eta'$ meson $|\eta_{M}| < 1$ and 0.9-2.1 $\mu$b for $2 < \eta_{M} < 5$. This opens the possibility to study diffractive production of pseudoscalar mesons in experiments at the LHC. We discuss if there are arguments from SU(3)-flavor symmetry which would forbid pomeron-pomeron fusion giving an $\eta$ meson. In our opinion such arguments do not exist. We also consider CEP of the pseudoscalars $\eta$ and $\eta'(958)$ and the pseudovector meson $f_{1}(1285)$ in diffractive proton-proton collisions in a theory with a scalar pomeron. We show that none of these particles can be produced in this way in the scalar-pomeron theory. Thus, experimental observation of any of these particles in the above CEP processes at the LHC would give striking evidence against a scalar character of the pomeron. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 10:14:26 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:54:23 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Lebiedowicz",
"Piotr",
""
],
[
"Nachtmann",
"Otto",
""
],
[
"Szczurek",
"Antoni",
""
]
] |
2506.04934 | Fabio Cavalletti | Fabio Cavalletti and Davide Manini and Andrea Mondino | On the geometry of synthetic null hypersurfaces | 52 pages | null | null | null | math.DG gr-qc math-ph math.MG math.MP | http://creativecommons.org/licenses/by/4.0/ | This paper develops a synthetic framework for the geometric and analytic study of null (lightlike) hypersurfaces in non-smooth spacetimes. Drawing from optimal transport and recent advances in Lorentzian geometry and causality theory, we define a \emph{synthetic null hypersurface} as a triple $(H, G, \mathfrak{m})$: $H$ is a closed achronal set in a topological causal space, $G$ is a gauge function encoding affine parametrizations along null generators, and $\mathfrak{m}$ is a Radon measure serving as a synthetic analog of the rigged measure. This generalizes classical differential geometric structures to potentially singular spacetimes.
A central object is the \emph{synthetic null energy condition} ($\mathsf{NC}^e(N)$), defined via the concavity of an entropy power functional along optimal transport, with parametrization given by the gauge $G$. This condition is invariant under changes of gauge and measure within natural equivalence classes. It agrees with the classical Null Energy Condition in the smooth setting and it applies to low-regularity spacetimes. A key property of ($\mathsf{NC}^e(N)$) is the stability under convergence of synthetic null hypersurfaces, inspired by measured Gromov--Hausdorff convergence.
As a first application, we obtain a synthetic version of Hawking's area theorem. Moreover, we extend the celebrated Penrose's singularity theorem to continuous spacetimes and prove the existence of trapped regions in the general setting of topological causal spaces satisfying the synthetic ($\mathsf{NC}^e(N)$). | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 12:09:09 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:03:07 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Cavalletti",
"Fabio",
""
],
[
"Manini",
"Davide",
""
],
[
"Mondino",
"Andrea",
""
]
] |
2506.05312 | Olaf D\"unkel | Olaf D\"unkel, Thomas Wimmer, Christian Theobalt, Christian Rupprecht, Adam Kortylewski | Do It Yourself: Learning Semantic Correspondence from Pseudo-Labels | Project page: https://genintel.github.io/DIY-SC | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Finding correspondences between semantically similar points across images and object instances is one of the everlasting challenges in computer vision. While large pre-trained vision models have recently been demonstrated as effective priors for semantic matching, they still suffer from ambiguities for symmetric objects or repeated object parts. We propose to improve semantic correspondence estimation via 3D-aware pseudo-labeling. Specifically, we train an adapter to refine off-the-shelf features using pseudo-labels obtained via 3D-aware chaining, filtering wrong labels through relaxed cyclic consistency, and 3D spherical prototype mapping constraints. While reducing the need for dataset specific annotations compared to prior work, we set a new state-of-the-art on SPair-71k by over 4% absolute gain and by over 7% against methods with similar supervision requirements. The generality of our proposed approach simplifies extension of training to other data sources, which we demonstrate in our experiments. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:54:33 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:30:41 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Dünkel",
"Olaf",
""
],
[
"Wimmer",
"Thomas",
""
],
[
"Theobalt",
"Christian",
""
],
[
"Rupprecht",
"Christian",
""
],
[
"Kortylewski",
"Adam",
""
]
] |
2506.05432 | Yuxuan Yue | Yuxuan Yue, Zukang Xu, Zhihang Yuan, Dawei Yang, Jianlong Wu, Liqiang Nie | PCDVQ: Enhancing Vector Quantization for Large Language Models via Polar Coordinate Decoupling | null | null | null | null | cs.LG cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) face significant challenges in edge deployment due to their massive parameter scale. Vector Quantization (VQ), a clustering-based quantization method, serves as a prevalent solution to this issue for its extremely low-bit (even at 2-bit) and considerable accuracy. Since a vector is a quantity in mathematics and physics that has both direction and magnitude, existing VQ works typically quantize them in a coupled manner. However, we find that direction exhibits significantly greater sensitivity to quantization compared to the magnitude. For instance, when separately clustering the directions and magnitudes of weight vectors in LLaMA-2-7B, the accuracy drop of zero-shot tasks are 46.5\% and 2.3\%, respectively. This gap even increases with the reduction of clustering centers. Further, Euclidean distance, a common metric to access vector similarities in current VQ works, places greater emphasis on reducing the magnitude error. This property is contrary to the above finding, unavoidably leading to larger quantization errors. To these ends, this paper proposes Polar Coordinate Decoupled Vector Quantization (PCDVQ), an effective and efficient VQ framework consisting of two key modules: 1) Polar Coordinate Decoupling (PCD), which transforms vectors into their polar coordinate representations and perform independent quantization of the direction and magnitude parameters.2) Distribution Aligned Codebook Construction (DACC), which optimizes the direction and magnitude codebooks in accordance with the source distribution. Experimental results show that PCDVQ outperforms baseline methods at 2-bit level by at least 1.5\% zero-shot accuracy, establishing a novel paradigm for accurate and highly compressed LLMs. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 08:58:58 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:17:49 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Yue",
"Yuxuan",
""
],
[
"Xu",
"Zukang",
""
],
[
"Yuan",
"Zhihang",
""
],
[
"Yang",
"Dawei",
""
],
[
"Wu",
"Jianlong",
""
],
[
"Nie",
"Liqiang",
""
]
] |
2506.05710 | Xiucheng Wang | Xiucheng Wang and Honggang Jia and Nan Cheng and Dusit Niyato | Latent Diffusion Model Based Denoising Receiver for 6G Semantic Communication: From Stochastic Differential Theory to Application | null | null | null | null | cs.LG cs.IT cs.SY eess.SY math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper, a novel semantic communication framework empowered by generative artificial intelligence (GAI) is proposed, to enhance the robustness against both channel noise and transmission data distribution shifts. A theoretical foundation is established using stochastic differential equations (SDEs), from which a closed-form mapping between any signal-to-noise ratio (SNR) and the optimal denoising timestep is derived. Moreover, to address distribution mismatch, a mathematical scaling method is introduced to align received semantic features with the training distribution of the GAI. Built on this theoretical foundation, a latent diffusion model (LDM)-based semantic communication framework is proposed that combines a variational autoencoder for semantic features extraction, where a pretrained diffusion model is used for denoising. The proposed system is a training-free framework that supports zero-shot generalization, and achieves superior performance under low-SNR and out-of-distribution conditions, offering a scalable and robust solution for future 6G semantic communication systems. Experimental results demonstrate that the proposed semantic communication framework achieves state-of-the-art performance in both pixel-level accuracy and semantic perceptual quality, consistently outperforming baselines across a wide range of SNRs and data distributions without any fine-tuning or post-training. | [
{
"version": "v1",
"created": "Fri, 06 Jun 2025 03:20:32 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 15:21:59 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Xiucheng",
""
],
[
"Jia",
"Honggang",
""
],
[
"Cheng",
"Nan",
""
],
[
"Niyato",
"Dusit",
""
]
] |
2506.06285 | Kaike Sa Teles Rocha Alves Mr. | Kaike Sa Teles Rocha Alves, Eduardo Pestana de Aguiar | NFISiS: New Perspectives on Fuzzy Inference Systems for Renewable Energy Forecasting | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Deep learning models, despite their popularity, face challenges such as long training times and a lack of interpretability. In contrast, fuzzy inference systems offer a balance of accuracy and transparency. This paper addresses the limitations of traditional Takagi-Sugeno-Kang fuzzy models by extending the recently proposed New Takagi-Sugeno-Kang model to a new Mamdani-based regressor. These models are data-driven, allowing users to define the number of rules to balance accuracy and interpretability. To handle the complexity of large datasets, this research integrates wrapper and ensemble techniques. A Genetic Algorithm is used as a wrapper for feature selection, creating genetic versions of the models. Furthermore, ensemble models, including the Random New Mamdani Regressor, Random New Takagi-Sugeno-Kang, and Random Forest New Takagi-Sugeno-Kang, are introduced to improve robustness. The proposed models are validated on photovoltaic energy forecasting datasets, a critical application due to the intermittent nature of solar power. Results demonstrate that the genetic and ensemble fuzzy models, particularly the Genetic New Takagi-Sugeno-Kang and Random Forest New Takagi-Sugeno-Kang, achieve superior performance. They often outperform both traditional machine learning and deep learning models while providing a simpler and more interpretable rule-based structure. The models are available online in a library called nfisis (https://pypi.org/project/nfisis/). | [
{
"version": "v1",
"created": "Mon, 28 Apr 2025 20:18:46 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 13:59:20 GMT"
},
{
"version": "v3",
"created": "Wed, 25 Jun 2025 22:00:25 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Alves",
"Kaike Sa Teles Rocha",
""
],
[
"de Aguiar",
"Eduardo Pestana",
""
]
] |
2506.06671 | Pedro Teles Dr | Pedro Teles | An analytical model for gold nanoparticle radiosensitisation | expanded Discussion, tweaked equations | null | null | null | physics.med-ph | http://creativecommons.org/licenses/by-sa/4.0/ | In this paper, we derive a variance-driven Local-Effect-Model ($\sigma$-LEM) to predict radiosensitization due to gold nanoparticles (AuNP). Assuming that the number of Au photo-ionisations scales strictly with particle volume $V_{\mathrm{NP}} \propto R^{3}$ and that the cascade energy deposition is log-normally distributed, the enhanced dose in each target voxel can be written as $D_{\text{enh}} = D_{0} \exp(\sigma Z)$ with $Z \sim \mathcal{N}(0,1)$ and width $\sigma = \sqrt{2 \ln(1 + Kc)}$.
By assuming a linear-quadratic (LQ) dose response, a relation between survival and dose can be derived for each voxel. Although there is no closed form for the log-normal distribution, averaging over the entire domain using first and second order moments leads to three possible closed forms: a variance-only one, a mixed-term one, where simultaneous hits from enhancement and baseline can occur, and a second-order model, accurate to $\mathcal{O}(\sigma^{3})$. These three variants adapt well to low-concentration, mid-concentration, and high-concentration regimes.
The model was tested for Bovine aortic endothelial cells (BAEC) results, irradiated with a 100 kVp x-ray beam, taken from a Local Effect Model (LEM) and experimental values. The model shows good agreement with the experimental data, with deviations within acceptable limits, but presents significant changes to the results obtained with the LEM. In particular, results seem to indicate that dose enhancement is mostly $\alpha$-driven. These findings are further developed in the manuscript.
The theoretical framework presented here collapses radiobiological outcomes to two experimentally controllable variables, beam quality, in the term $K_c$, and intracellular concentration $c$, while retaining mechanistic fidelity. Additional tests should be made to further confirm the validity of the model. | [
{
"version": "v1",
"created": "Sat, 07 Jun 2025 05:28:46 GMT"
},
{
"version": "v2",
"created": "Wed, 11 Jun 2025 02:12:03 GMT"
},
{
"version": "v3",
"created": "Wed, 25 Jun 2025 03:36:49 GMT"
},
{
"version": "v4",
"created": "Thu, 26 Jun 2025 05:41:51 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Teles",
"Pedro",
""
]
] |
2506.07123 | Mahdi Bashiri Bawil | Mahdi Bashiri Bawil, Mousa Shamsi, Abolhassan Shakeri Bavil | Simultaneous Segmentation of Ventricles and Normal/Abnormal White Matter Hyperintensities in Clinical MRI using Deep Learning | 43 pages, 11 figures, 1 table | null | null | null | eess.IV | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Multiple sclerosis (MS) diagnosis and monitoring rely heavily on accurate assessment of brain MRI biomarkers, particularly white matter hyperintensities (WMHs) and ventricular changes. Current segmentation approaches suffer from several limitations: they typically segment these structures independently despite their pathophysiological relationship, struggle to differentiate between normal and pathological hyperintensities, and are poorly optimized for anisotropic clinical MRI data. We propose a novel 2D pix2pix-based deep learning framework for simultaneous segmentation of ventricles and WMHs with the unique capability to distinguish between normal periventricular hyperintensities and pathological MS lesions. Our method was developed and validated on FLAIR MRI scans from 300 MS patients. Compared to established methods (SynthSeg, Atlas Matching, BIANCA, LST-LPA, LST-LGA, and WMH-SynthSeg), our approach achieved superior performance for both ventricle segmentation (Dice: 0.801+/-0.025, HD95: 18.46+/-7.1mm) and WMH segmentation (Dice: 0.624+/-0.061, precision: 0.755+/-0.161). Furthermore, our method successfully differentiated between normal and abnormal hyperintensities with a Dice coefficient of 0.647. Notably, our approach demonstrated exceptional computational efficiency, completing end-to-end processing in approximately 4 seconds per case, up to 36 times faster than baseline methods, while maintaining minimal resource requirements. This combination of improved accuracy, clinically relevant differentiation capability, and computational efficiency addresses critical limitations in current neuroimaging analysis, potentially enabling integration into routine clinical workflows and enhancing MS diagnosis and monitoring. | [
{
"version": "v1",
"created": "Sun, 08 Jun 2025 13:09:51 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 22:08:48 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Bawil",
"Mahdi Bashiri",
""
],
[
"Shamsi",
"Mousa",
""
],
[
"Bavil",
"Abolhassan Shakeri",
""
]
] |
2506.07413 | Ziwen Wang | Ziwen Wang, Jiajun Fan, Thao Nguyen, Heng Ji, Ge Liu | Variational Supervised Contrastive Learning | null | null | null | null | cs.LG cs.CV | http://creativecommons.org/licenses/by/4.0/ | Contrastive learning has proven to be highly efficient and adaptable in shaping representation spaces across diverse modalities by pulling similar samples together and pushing dissimilar ones apart. However, two key limitations persist: (1) Without explicit regulation of the embedding distribution, semantically related instances can inadvertently be pushed apart unless complementary signals guide pair selection, and (2) excessive reliance on large in-batch negatives and tailored augmentations hinders generalization. To address these limitations, we propose Variational Supervised Contrastive Learning (VarCon), which reformulates supervised contrastive learning as variational inference over latent class variables and maximizes a posterior-weighted evidence lower bound (ELBO) that replaces exhaustive pair-wise comparisons for efficient class-aware matching and grants fine-grained control over intra-class dispersion in the embedding space. Trained exclusively on image data, our experiments on CIFAR-10, CIFAR-100, ImageNet-100, and ImageNet-1K show that VarCon (1) achieves state-of-the-art performance for contrastive learning frameworks, reaching 79.36% Top-1 accuracy on ImageNet-1K and 78.29% on CIFAR-100 with a ResNet-50 encoder while converging in just 200 epochs; (2) yields substantially clearer decision boundaries and semantic organization in the embedding space, as evidenced by KNN classification, hierarchical clustering results, and transfer-learning assessments; and (3) demonstrates superior performance in few-shot learning than supervised baseline and superior robustness across various augmentation strategies. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 04:19:12 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:27:25 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Ziwen",
""
],
[
"Fan",
"Jiajun",
""
],
[
"Nguyen",
"Thao",
""
],
[
"Ji",
"Heng",
""
],
[
"Liu",
"Ge",
""
]
] |
2506.07537 | Jeongmin Han | Jeongmin Han | Tug-of-war games related to $p$-Laplace type equations with lower order terms | null | null | null | null | math.AP | http://creativecommons.org/licenses/by/4.0/ | In this paper, we investigate a class of tug-of-war games that incorporate a constant payoff discount rate at each turn. The associated model problems are $p$-Laplace type partial differential equations with zeroth-order terms. We establish existence, uniqueness, and regularity results for the corresponding game value functions. Furthermore, we explore properties of the solutions to the model PDEs, informed by the analysis of the underlying games. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 08:18:39 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 03:38:06 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Han",
"Jeongmin",
""
]
] |
2506.07625 | Steven Finch | Steven Finch | Half-Iterates of $x$exp$(x)$, $x+1/x$ and arcsinh$(x)$ | A distinct preprint covering half-iterates of $x(1+x)$, $\sin(x)$ and $\exp(x/e)$ appears at arXiv:2506.07625v1. Reading this earlier work will help to motivate the study of $\delta$ in the current paper. 15 pages; 4 figures | null | null | null | math.NT cs.DM | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Given $\theta(x)$, one method (EJ) for solving Abel's equation $g(\theta(x)) = g(x) \pm 1$ is significantly faster than a rival method (ML). On the other hand, ML evaluates a limit characterizing the principal solution $g(x)$ directly while EJ finds $g(x) + \delta$, where $\delta$ is possibly nonzero but independent of $x$. If an exact expression for $\delta$ is known, then the "intrinsicality" of ML carries over and relative quickness of EJ is preserved. We study $\delta$, as determined by $\theta$, and continue tangentially our earlier exploration of compositional square roots. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 10:42:45 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 02:12:58 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Finch",
"Steven",
""
]
] |
2506.07640 | Ruopengyu Xu | Ruopengyu Xu and Chenglian Liu | Stark-Coleman Invariants and Quantum Lower Bounds: An Integrated Framework for Real Quadratic Fields | 16 pages, 1 figure, 3 tables | null | null | null | math.NT cs.CR math.GR quant-ph | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Class groups of real quadratic fields represent fundamental structures in algebraic number theory with significant computational implications. While Stark's conjecture establishes theoretical connections between special units and class group structures, explicit constructions have remained elusive, and precise quantum complexity bounds for class group computations are lacking. Here we establish an integrated framework defining Stark-Coleman invariants $\kappa_p(K) = \log_p \left( \frac{\varepsilon_{\mathrm{St},p}}{\sigma(\varepsilon_{\mathrm{St},p})} \right) \mod p^{\mathrm{ord}_p(\Delta_K)}$ through a synthesis of $p$-adic Hodge theory and extended Coleman integration. We prove these invariants classify class groups under the Generalized Riemann Hypothesis (GRH), resolving the isomorphism problem for discriminants $D > 10^{32}$. Furthermore, we demonstrate that this approach yields the quantum lower bound $\exp\left(\Omega\left(\frac{\log D}{(\log \log D)^2}\right)\right)$ for the class group discrete logarithm problem, improving upon previous bounds lacking explicit constants. Our results indicate that Stark units constrain the geometric organization of class groups, providing theoretical insight into computational complexity barriers. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 11:06:17 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Xu",
"Ruopengyu",
""
],
[
"Liu",
"Chenglian",
""
]
] |
2506.07886 | Gen Li | Gen Li, Yutong Chen, Yiqian Wu, Kaifeng Zhao, Marc Pollefeys, Siyu Tang | EgoM2P: Egocentric Multimodal Multitask Pretraining | Accepted by ICCV 2025 | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Understanding multimodal signals in egocentric vision, such as RGB video, depth, camera poses, and gaze, is essential for applications in augmented reality, robotics, and human-computer interaction, enabling systems to better interpret the camera wearer's actions, intentions, and surrounding environment. However, building large-scale egocentric multimodal and multitask models presents unique challenges. Egocentric data are inherently heterogeneous, with large variations in modality coverage across devices and settings. Generating pseudo-labels for missing modalities, such as gaze or head-mounted camera trajectories, is often infeasible, making standard supervised learning approaches difficult to scale. Furthermore, dynamic camera motion and the complex temporal and spatial structure of first-person video pose additional challenges for the direct application of existing multimodal foundation models.
To address these challenges, we introduce a set of efficient temporal tokenizers and propose EgoM2P, a masked modeling framework that learns from temporally-aware multimodal tokens to train a large, general-purpose model for egocentric 4D understanding. This unified design supports multitasking across diverse egocentric perception and synthesis tasks, including gaze prediction, egocentric camera tracking, and monocular depth estimation from egocentric video, and also serves as a generative model for conditional egocentric video synthesis. Across these tasks, EgoM2P matches or outperforms specialist models while being an order of magnitude faster. We will fully open-source EgoM2P to support the community and advance egocentric vision research. Project page: https://egom2p.github.io/. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 15:59:25 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:13:31 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Li",
"Gen",
""
],
[
"Chen",
"Yutong",
""
],
[
"Wu",
"Yiqian",
""
],
[
"Zhao",
"Kaifeng",
""
],
[
"Pollefeys",
"Marc",
""
],
[
"Tang",
"Siyu",
""
]
] |
2506.07977 | Jingjing Chang | Jingjing Chang, Yixiao Fang, Peng Xing, Shuhan Wu, Wei Cheng, Rui Wang, Xianfang Zeng, Gang Yu, Hai-Bao Chen | OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Text-to-image (T2I) models have garnered significant attention for generating high-quality images aligned with text prompts. However, rapid T2I model advancements reveal limitations in early benchmarks, lacking comprehensive evaluations, for example, the evaluation on reasoning, text rendering and style. Notably, recent state-of-the-art models, with their rich knowledge modeling capabilities, show promising results on the image generation problems requiring strong reasoning ability, yet existing evaluation systems have not adequately addressed this frontier. To systematically address these gaps, we introduce OneIG-Bench, a meticulously designed comprehensive benchmark framework for fine-grained evaluation of T2I models across multiple dimensions, including prompt-image alignment, text rendering precision, reasoning-generated content, stylization, and diversity. By structuring the evaluation, this benchmark enables in-depth analysis of model performance, helping researchers and practitioners pinpoint strengths and bottlenecks in the full pipeline of image generation. Specifically, OneIG-Bench enables flexible evaluation by allowing users to focus on a particular evaluation subset. Instead of generating images for the entire set of prompts, users can generate images only for the prompts associated with the selected dimension and complete the corresponding evaluation accordingly. Our codebase and dataset are now publicly available to facilitate reproducible evaluation studies and cross-model comparisons within the T2I research community. | [
{
"version": "v1",
"created": "Mon, 09 Jun 2025 17:50:21 GMT"
},
{
"version": "v2",
"created": "Tue, 10 Jun 2025 12:31:25 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 15:47:09 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Chang",
"Jingjing",
""
],
[
"Fang",
"Yixiao",
""
],
[
"Xing",
"Peng",
""
],
[
"Wu",
"Shuhan",
""
],
[
"Cheng",
"Wei",
""
],
[
"Wang",
"Rui",
""
],
[
"Zeng",
"Xianfang",
""
],
[
"Yu",
"Gang",
""
],
[
"Chen",
"Hai-Bao",
""
]
] |
2506.08940 | Adam Bednorz | Adam Bednorz, Josep Batle, Tomasz Bia{\l}ecki, Jaros{\l}aw K. Korbicz | Closing objectivity loophole in Bell tests on a public quantum computer | 13 pages, 12 figures, 7 tables | null | null | null | quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We have constructed and run a Bell test of local realism focusing on the objectivity criterion. The objectivity means that the outcomes are confirmed macroscopically by a few observers at each party. The IBM Quantum and IonQ devices turn out to be sufficiently accurate to pass such an extended Bell-type test, although at the price of communication loopholes and residual but statistically significant signaling. The test also serves as the benchmark of entanglement spread across larger sets of qubits. | [
{
"version": "v1",
"created": "Tue, 10 Jun 2025 16:03:39 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:15:13 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Bednorz",
"Adam",
""
],
[
"Batle",
"Josep",
""
],
[
"Białecki",
"Tomasz",
""
],
[
"Korbicz",
"Jarosław K.",
""
]
] |
2506.09006 | Harri Waltari | Stefano Moretti, Luca Panizzi, J\"orgen Sj\"olin and Harri Waltari | Deconstructing resonant Higgs pair production at the LHC: effects of coloured and neutral scalars in the NMSSM test case | 20 pages, 9 figures, a new appendix on the UFO model and a link to it, also some added references | null | null | null | hep-ph hep-ex | http://creativecommons.org/licenses/by-nc-sa/4.0/ | We study resonant production of pairs of Standard Model (SM)-like Higgs bosons, in the presence of new neutral Higgs states together with new coloured scalars (stops or sbottoms) in loops within the Next-to-Minimal Supersymmetric SM (NMSSM). This is used as a test case to prove that the Large Hadron Collider has sensitivity to a variety of effects stemming from interferences between resonant (heavy) Higgs diagrams and/or among these and non-resonant topologies involving loops of both tops and stops. These effects can alter significantly the naive description of individual $s$-channel Breit-Wigner resonances, leading to distortions of the latter which, on the one hand, may mask their presence but, on the other hand, could enable one to extract features of the underlying new physics scenario. This last aspect is made possible through a decomposition of the $gg\to hh$ signal process into all its amplitude components, each of which has a well-defined coupling structure. Ultimately, such effects can be traced back to the relevant Feynman diagrams and can enable a detailed interpretation of this process. To illustrate this, we introduce various Benchmark Points that exhibit potentially observable features during the current and/or upcoming runs of the LHC in one or more of the three customary di-Higgs decay channels: $b\bar bb\bar b$, $b\bar b \tau^+\tau^-$ and $b\bar b\gamma\gamma$. | [
{
"version": "v1",
"created": "Tue, 10 Jun 2025 17:27:49 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:01:08 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Moretti",
"Stefano",
""
],
[
"Panizzi",
"Luca",
""
],
[
"Sjölin",
"Jörgen",
""
],
[
"Waltari",
"Harri",
""
]
] |
2506.09498 | Hyeonseo Cho | Jaesik Yoon, Hyeonseo Cho, Yoshua Bengio, Sungjin Ahn | Fast Monte Carlo Tree Diffusion: 100x Speedup via Parallel Sparse Planning | null | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | Diffusion models have recently emerged as a powerful approach for trajectory planning. However, their inherently non-sequential nature limits their effectiveness in long-horizon reasoning tasks at test time. The recently proposed Monte Carlo Tree Diffusion (MCTD) offers a promising solution by combining diffusion with tree-based search, achieving state-of-the-art performance on complex planning problems. Despite its strengths, our analysis shows that MCTD incurs substantial computational overhead due to the sequential nature of tree search and the cost of iterative denoising. To address this, we propose Fast-MCTD, a more efficient variant that preserves the strengths of MCTD while significantly improving its speed and scalability. Fast-MCTD integrates two techniques: Parallel MCTD, which enables parallel rollouts via delayed tree updates and redundancy-aware selection; and Sparse MCTD, which reduces rollout length through trajectory coarsening. Experiments show that Fast-MCTD achieves up to 100x speedup over standard MCTD while maintaining or improving planning performance. Remarkably, it even outperforms Diffuser in inference speed on some tasks, despite Diffuser requiring no search and yielding weaker solutions. These results position Fast-MCTD as a practical and scalable solution for diffusion-based inference-time reasoning. | [
{
"version": "v1",
"created": "Wed, 11 Jun 2025 08:17:40 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 01:52:43 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Yoon",
"Jaesik",
""
],
[
"Cho",
"Hyeonseo",
""
],
[
"Bengio",
"Yoshua",
""
],
[
"Ahn",
"Sungjin",
""
]
] |
2506.09579 | Pengfei Wang | Pengfei Wang, Ziyang Zhang, Wensong Wang, Shuangmin Chen, Lin Lu, Shiqing Xin, Changhe Tu | Power Diagram Enhanced Adaptive Isosurface Extraction from Signed Distance Fields | null | null | null | null | cs.CG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Extracting high-fidelity mesh surfaces from Signed Distance Fields has become a fundamental operation in geometry processing. Despite significant progress over the past decades, key challenges remain namely, how to automatically capture the intricate geometric and topological structures encoded in the zero level set of SDFs. In this paper, we present a novel isosurface extraction algorithm that introduces two key innovations: 1. An incrementally constructed power diagram through the addition of sample points, which enables repeated updates to the extracted surface via its dual regular Delaunay tetrahedralization; and 2. An adaptive point insertion strategy that identifies regions exhibiting the greatest discrepancy between the current mesh and the underlying continuous surface. As the teaser figure shows, our framework progressively refines the extracted mesh with minimal computational cost until it sufficiently approximates the underlying surface. Experimental results demonstrate that our approach outperforms sofa methods, particularly for models with intricate geometric variations and complex topologies. | [
{
"version": "v1",
"created": "Wed, 11 Jun 2025 10:17:36 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:19:48 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Pengfei",
""
],
[
"Zhang",
"Ziyang",
""
],
[
"Wang",
"Wensong",
""
],
[
"Chen",
"Shuangmin",
""
],
[
"Lu",
"Lin",
""
],
[
"Xin",
"Shiqing",
""
],
[
"Tu",
"Changhe",
""
]
] |
2506.09803 | Longzhu He | Longzhu He, Chaozhuo Li, Peng Tang, Li Sun, Sen Su, Philip S. Yu | Devil's Hand: Data Poisoning Attacks to Locally Private Graph Learning Protocols | null | null | null | null | cs.LG cs.CR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Graph neural networks (GNNs) have achieved significant success in graph representation learning and have been applied to various domains. However, many real-world graphs contain sensitive personal information, such as user profiles in social networks, raising serious privacy concerns when graph learning is performed using GNNs. To address this issue, locally private graph learning protocols have gained considerable attention. These protocols leverage the privacy advantages of local differential privacy (LDP) and the effectiveness of GNN's message-passing in calibrating noisy data, offering strict privacy guarantees for users' local data while maintaining high utility (e.g., node classification accuracy) for graph learning. Despite these advantages, such protocols may be vulnerable to data poisoning attacks, a threat that has not been considered in previous research. Identifying and addressing these threats is crucial for ensuring the robustness and security of privacy-preserving graph learning frameworks. This work introduces the first data poisoning attack targeting locally private graph learning protocols. The attacker injects fake users into the protocol, manipulates these fake users to establish links with genuine users, and sends carefully crafted data to the server, ultimately compromising the utility of private graph learning. The effectiveness of the attack is demonstrated both theoretically and empirically. In addition, several defense strategies have also been explored, but their limited effectiveness highlights the need for more robust defenses. | [
{
"version": "v1",
"created": "Wed, 11 Jun 2025 14:46:11 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:18:21 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"He",
"Longzhu",
""
],
[
"Li",
"Chaozhuo",
""
],
[
"Tang",
"Peng",
""
],
[
"Sun",
"Li",
""
],
[
"Su",
"Sen",
""
],
[
"Yu",
"Philip S.",
""
]
] |
2506.09850 | Henrique Bolfarine | Henrique Bolfarine, Hedibert F. Lopes, Carlos M. Carvalho | Lower-dimensional posterior density and cluster summaries for overparameterized Bayesian models | 50 pages, 9 Figures | null | null | null | stat.ME | http://creativecommons.org/licenses/by/4.0/ | The usefulness of Bayesian models for density and cluster estimation is well established across multiple literatures. However, there is still a known tension between the use of simpler, more interpretable models and more flexible, complex ones. In this paper, we propose a novel method that integrates these two approaches by projecting the fit of a flexible, over-parameterized model onto a lower-dimensional parametric summary, which serves as a surrogate. This process increases interpretability while preserving most of the fit of the original model. Our approach involves three main steps. First, we fit the data using nonparametric or over-parameterized models. Second, we project the posterior predictive distribution of the original model onto a sequence of parametric summary estimates using a decision-theoretic approach. Finally, given the lower parametric dimension of the summary estimate that best approximates the original model learned in the second step, we construct uncertainty quantification for the summary by projecting the original full posterior distribution. We demonstrate the effectiveness of our method in summarizing a variety of nonparametric and overparameterized models, providing uncertainty quantification for both density and cluster summaries on synthetic and real datasets. | [
{
"version": "v1",
"created": "Wed, 11 Jun 2025 15:21:57 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:47:32 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Bolfarine",
"Henrique",
""
],
[
"Lopes",
"Hedibert F.",
""
],
[
"Carvalho",
"Carlos M.",
""
]
] |
2506.10640 | Daniel Levy | Daniel Levy, Haipeng (Allan) Chen, Sourav Ray, Elliot Charette, Xiao Ling, Weihong Zhao, Mark Bergen, and Avichai Snir | Asymmetric price adjustment over the business cycle | null | Economics Letters (Forthcoming), 2025 | null | null | econ.GN q-fin.EC | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Studies of micro-level price datasets find more frequent small price increases than decreases, which can be explained by consumer inattention because time-constrained shoppers might ignore small price changes. Recent empirical studies of the link between shopping behavior and price attention over the business cycle find that consumers are more attentive to prices during economic downturns, and less attentive during economic booms. These two sets of findings have a testable implication. The asymmetry in small price changes should vary over the business cycle. It should diminish during recessions and strengthen during expansions. We test this prediction using a large US store-level dataset with more than 98 million weekly price observations for the years 1989-1997, which includes an 8-month recession period, as defined by the NBER. We compare price adjustments between periods of recession - high unemployment, and expansion - low unemployment. Focusing on small price changes, we find, consistent with our hypothesis, that there is a greater asymmetry in small price changes during periods of low unemployment compared to the periods of high unemployment, implying that firms price-setting behavior varies over the business cycle. | [
{
"version": "v1",
"created": "Thu, 12 Jun 2025 12:25:47 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:46:03 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Levy",
"Daniel",
"",
"Allan"
],
[
"Haipeng",
"",
"",
"Allan"
],
[
"Chen",
"",
""
],
[
"Ray",
"Sourav",
""
],
[
"Charette",
"Elliot",
""
],
[
"Ling",
"Xiao",
""
],
[
"Zhao",
"Weihong",
""
],
[
"Bergen",
"Mark",
""
],
[
"Snir",
"Avichai",
""
]
] |
2506.11924 | Min-Seop Kwak | Min-Seop Kwak, Junho Kim, Sangdoo Yun, Dongyoon Han, Taekyoung Kim, Seungryong Kim, Jin-Hwa Kim | Aligned Novel View Image and Geometry Synthesis via Cross-modal Attention Instillation | Project page at https://cvlab-kaist.github.io/MoAI | null | null | null | cs.CV | http://creativecommons.org/licenses/by-nc-nd/4.0/ | We introduce a diffusion-based framework that performs aligned novel view image and geometry generation via a warping-and-inpainting methodology. Unlike prior methods that require dense posed images or pose-embedded generative models limited to in-domain views, our method leverages off-the-shelf geometry predictors to predict partial geometries viewed from reference images, and formulates novel-view synthesis as an inpainting task for both image and geometry. To ensure accurate alignment between generated images and geometry, we propose cross-modal attention distillation, where attention maps from the image diffusion branch are injected into a parallel geometry diffusion branch during both training and inference. This multi-task approach achieves synergistic effects, facilitating geometrically robust image synthesis as well as well-defined geometry prediction. We further introduce proximity-based mesh conditioning to integrate depth and normal cues, interpolating between point cloud and filtering erroneously predicted geometry from influencing the generation process. Empirically, our method achieves high-fidelity extrapolative view synthesis on both image and geometry across a range of unseen scenes, delivers competitive reconstruction quality under interpolation settings, and produces geometrically aligned colored point clouds for comprehensive 3D completion. Project page is available at https://cvlab-kaist.github.io/MoAI. | [
{
"version": "v1",
"created": "Fri, 13 Jun 2025 16:19:00 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 15:26:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kwak",
"Min-Seop",
""
],
[
"Kim",
"Junho",
""
],
[
"Yun",
"Sangdoo",
""
],
[
"Han",
"Dongyoon",
""
],
[
"Kim",
"Taekyoung",
""
],
[
"Kim",
"Seungryong",
""
],
[
"Kim",
"Jin-Hwa",
""
]
] |
2506.12025 | Sonia Mazelet | Sonia Mazelet, R\'emi Flamary, Bertrand Thirion | Unsupervised Learning for Optimal Transport plan prediction between unbalanced graphs | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Optimal transport between graphs, based on Gromov-Wasserstein and
other extensions, is a powerful tool for comparing and aligning
graph structures. However, solving the associated non-convex
optimization problems is computationally expensive, which limits the
scalability of these methods to large graphs. In this work, we
present Unbalanced Learning of Optimal Transport (ULOT), a deep
learning method that predicts optimal transport plans between two
graphs. Our method is trained by minimizing the fused unbalanced
Gromov-Wasserstein (FUGW) loss. We propose a novel neural
architecture with cross-attention that is conditioned on the FUGW
tradeoff hyperparameters. We evaluate ULOT on synthetic stochastic
block model (SBM) graphs and on real cortical surface data obtained
from fMRI. ULOT predicts transport plans with competitive loss up to
two orders of magnitude faster than classical solvers. Furthermore,
the predicted plan can be used as a warm start for classical solvers
to accelerate their convergence. Finally, the predicted transport
plan is fully differentiable with respect to the graph inputs and
FUGW hyperparameters, enabling the optimization of functionals of
the ULOT plan. | [
{
"version": "v1",
"created": "Wed, 21 May 2025 09:29:19 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:01:32 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Mazelet",
"Sonia",
""
],
[
"Flamary",
"Rémi",
""
],
[
"Thirion",
"Bertrand",
""
]
] |
2506.12113 | Gr\'egoire Barru\'e | Benjamin Marais, Tony Quertier, Gr\'egoire Barrue | Semantic Preprocessing for LLM-based Malware Analysis | null | null | null | null | cs.CR cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In a context of malware analysis, numerous approaches rely on Artificial Intelligence to handle a large volume of data. However, these techniques focus on data view (images, sequences) and not on an expert's view. Noticing this issue, we propose a preprocessing that focuses on expert knowledge to improve malware semantic analysis and result interpretability. We propose a new preprocessing method which creates JSON reports for Portable Executable files. These reports gather features from both static and behavioral analysis, and incorporate packer signature detection, MITRE ATT\&CK and Malware Behavior Catalog (MBC) knowledge. The purpose of this preprocessing is to gather a semantic representation of binary files, understandable by malware analysts, and that can enhance AI models' explainability for malicious files analysis. Using this preprocessing to train a Large Language Model for Malware classification, we achieve a weighted-average F1-score of 0.94 on a complex dataset, representative of market reality. | [
{
"version": "v1",
"created": "Fri, 13 Jun 2025 13:39:00 GMT"
},
{
"version": "v2",
"created": "Thu, 19 Jun 2025 09:55:01 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 15:09:42 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Marais",
"Benjamin",
""
],
[
"Quertier",
"Tony",
""
],
[
"Barrue",
"Grégoire",
""
]
] |
2506.12162 | Lawrence Hollom | Lawrence Hollom | Finding long cycles in a percolated expander graphs | 11 pages | null | null | null | math.CO math.PR | http://creativecommons.org/licenses/by/4.0/ | Given a graph $G$, the percolated graph $G_p$ has each edge independently retained with probability $p$. Collares, Diskin, Erde, and Krivelevich initiated the study of large structures in percolated single-scale vertex expander graphs, wherein every set of exactly $k$ vertices of $G$ has at least $dk$ neighbours before percolation. We extend their result to a conjectured stronger form, proving that if $p = (1+\varepsilon)/d$ and $G$ is a graph on at least $k$ vertices which expands as above, then $G_p$ contains a cycle of length $\Omega_\varepsilon(kd)$ with probability at least $1-\exp(-\Omega_\varepsilon(k/d))$ as $k\rightarrow\infty$. | [
{
"version": "v1",
"created": "Fri, 13 Jun 2025 18:28:40 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:24:27 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Hollom",
"Lawrence",
""
]
] |
2506.12256 | Anita Varga | Joonyeob Lee, D\'avid Papp, Anita Varga | Dual certificates of primal cone membership | 24 pages; Clarified a notation and fixed minor typographical errors | null | null | null | math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We discuss easily verifiable cone membership certificates, that is, certificates proving relations of the form $b\in K$ for convex cones $K$, that consist of vectors in the dual cone $K^*$. Vectors in the dual cone are usually associated with separating hyperplanes, and so they are interpreted as certificates of non-membership in the standard theory of duality. Complementing this, we present constructive certification schemes through which members of the dual cone can be interpreted as primal membership certificates. Every vector in the interior of $K$ is assigned a full-dimensional cone of certificates, making the numerical computation of rigorous certificates easy, provided that the dual cone has an efficiently computable logarithmically homogeneous self-concordant barrier. Of particular interest are cones that are low-dimensional linear images of much higher dimensional cones. In the context of optimization (as opposed to feasibility) problems, these certificates can be used to easily compute, using a closed-form formula, exact primal feasible solutions from suitable dual feasible solutions, with the guarantee that the closer the dual solutions are to optimality, the closer to optimality are the computed primal solutions, too. We demonstrate that the new certification scheme is applicable to virtually every tractable subcone of nonnegative polynomials commonly used in polynomial optimization (such as SOS, SONC, SAGE and SDSOS polynomials, among others), facilitating the computation of rigorous nonnegativity certificates using numerical algorithms. | [
{
"version": "v1",
"created": "Fri, 13 Jun 2025 22:10:46 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:37:58 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Lee",
"Joonyeob",
""
],
[
"Papp",
"Dávid",
""
],
[
"Varga",
"Anita",
""
]
] |
2506.12896 | Taiga Hayami | Taiga Hayami, Kakeru Koizumi, Hiroshi Watanabe | Structure-Preserving Patch Decoding for Efficient Neural Video Representation | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Implicit neural representations (INRs) are the subject of extensive research, particularly in their application to modeling complex signals by mapping spatial and temporal coordinates to corresponding values. When handling videos, mapping compact inputs to entire frames or spatially partitioned patch images is an effective approach. This strategy better preserves spatial relationships, reduces computational overhead, and improves reconstruction quality compared to coordinate-based mapping. However, predicting entire frames often limits the reconstruction of high-frequency visual details. Additionally, conventional patch-based approaches based on uniform spatial partitioning tend to introduce boundary discontinuities that degrade spatial coherence. We propose a neural video representation method based on Structure-Preserving Patches (SPPs) to address such limitations. Our method separates each video frame into patch images of spatially aligned frames through a deterministic pixel-based splitting similar to PixelUnshuffle. This operation preserves the global spatial structure while allowing patch-level decoding. We train the decoder to reconstruct these structured patches, enabling a global-to-local decoding strategy that captures the global layout first and refines local details. This effectively reduces boundary artifacts and mitigates distortions from naive upsampling. Experiments on standard video datasets demonstrate that our method achieves higher reconstruction quality and better compression performance than existing INR-based baselines. | [
{
"version": "v1",
"created": "Sun, 15 Jun 2025 15:58:23 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:27:09 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Hayami",
"Taiga",
""
],
[
"Koizumi",
"Kakeru",
""
],
[
"Watanabe",
"Hiroshi",
""
]
] |
2506.13056 | Haibo Qiu | Haibo Qiu, Xiaohan Lan, Fanfan Liu, Xiaohu Sun, Delian Ruan, Peng Shi, Lin Ma | Metis-RISE: RL Incentivizes and SFT Enhances Multimodal Reasoning Model Learning | Project Page: https://github.com/MM-Thinking/Metis-RISE | null | null | null | cs.AI cs.CV cs.LG | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Recent advancements in large language models (LLMs) have witnessed a surge in the development of advanced reasoning paradigms, which are now being integrated into multimodal large language models (MLLMs). However, existing approaches often fall short: methods solely employing reinforcement learning (RL) can struggle with sample inefficiency and activating entirely absent reasoning capabilities, while conventional pipelines that initiate with a cold-start supervised fine-tuning (SFT) phase before RL may restrict the model's exploratory capacity and face suboptimal convergence. In this work, we introduce \textbf{Metis-RISE} (\textbf{R}L \textbf{I}ncentivizes and \textbf{S}FT \textbf{E}nhances) for multimodal reasoning model learning. Unlike conventional approaches, Metis-RISE distinctively omits an initial SFT stage, beginning instead with an RL phase (e.g., using a Group Relative Policy Optimization variant) to incentivize and activate the model's latent reasoning capacity. Subsequently, the targeted SFT stage addresses two key challenges identified during RL: (1) \textit{inefficient trajectory sampling} for tasks where the model possesses but inconsistently applies correct reasoning, which we tackle using self-distilled reasoning trajectories from the RL model itself; and (2) \textit{fundamental capability absence}, which we address by injecting expert-augmented knowledge for prompts where the model entirely fails. This strategic application of RL for incentivization followed by SFT for enhancement forms the core of Metis-RISE, leading to two versions of our MLLMs (7B and 72B parameters). Evaluations on the OpenCompass Multimodal Reasoning Leaderboard demonstrate that both models achieve state-of-the-art performance among similar-sized models, with the 72B version ranking fourth overall. Please refer to our project page for open-source information. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 02:56:13 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 11:45:11 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Qiu",
"Haibo",
""
],
[
"Lan",
"Xiaohan",
""
],
[
"Liu",
"Fanfan",
""
],
[
"Sun",
"Xiaohu",
""
],
[
"Ruan",
"Delian",
""
],
[
"Shi",
"Peng",
""
],
[
"Ma",
"Lin",
""
]
] |
2506.13242 | Alexandre Sedoglavic | Jean-Guillaume Dumas (CASC), Cl\'ement Pernet (CASC), Alexandre Sedoglavic (CRIStAL) | A non-commutative algorithm for multiplying 4x4 matrices using 48 non-complex multiplications | null | null | null | null | cs.SC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The quest for non-commutative matrix multiplication algorithms in small dimensions has seen a lot of recent improvements recently. In particular, the number of scalar multiplications required to multiply two $4\times4$ matrices was first reduced in \cite{Fawzi:2022aa} from 49 (two recursion levels of Strassen's algorithm) to 47 but only in characteristic 2 or more recently to 48 in \cite{alphaevolve} but over complex numbers. We propose an algorithm in 48 multiplications with only rational coefficients, hence removing the complex number requirement. It was derived from the latter one, under the action of an isotropy which happen to project the algorithm on the field of rational numbers. We also produce a straight line program of this algorithm, reducing the leading constant in the complexity, as well as an alternative basis variant of it, leading to an algorithm running in $\frac{19}{16} n^{2+\frac{\log_2 3}{2}} +o\left(n^{2+\frac{log_2 3}{2}}\right)$ operations over any ring containing an inverse of 2. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 08:42:15 GMT"
},
{
"version": "v2",
"created": "Tue, 17 Jun 2025 08:51:21 GMT"
},
{
"version": "v3",
"created": "Tue, 24 Jun 2025 08:34:48 GMT"
},
{
"version": "v4",
"created": "Thu, 26 Jun 2025 09:00:56 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Dumas",
"Jean-Guillaume",
"",
"CASC"
],
[
"Pernet",
"Clément",
"",
"CASC"
],
[
"Sedoglavic",
"Alexandre",
"",
"CRIStAL"
]
] |
2506.13512 | David P. Blecher | David P. Blecher and Caleb McClure | Real Non-Commutative Convexity I | 40 pages | null | null | null | math.OA math-ph math.FA math.MP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We initiate the theory of real noncommutative (nc) convex sets, the real case of the recent and profound complex theory developed by Davidson and Kennedy. The present paper focuses on the real case of the topics from the first several sections of their profound memoir. Later results will be discussed in a future paper. We develop some of the infrastructure of real nc convexity, giving many foundational structural results for real operator systems and their associated nc convex sets, and elucidate how the complexification interacts with the basic convexity theory constructions. Several new features appear in the real case, including the novel notion of the complexification of a nc convex set. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 14:07:56 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:39:02 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Blecher",
"David P.",
""
],
[
"McClure",
"Caleb",
""
]
] |
2506.13846 | Runtao Liu | Runtao Liu, Jiahao Zhan, Yingqing He, Chen Wei, Alan Yuille, Qifeng Chen | Fake it till You Make it: Reward Modeling as Discriminative Prediction | null | null | null | null | cs.CV cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | An effective reward model plays a pivotal role in reinforcement learning for post-training enhancement of visual generative models. However, current approaches of reward modeling suffer from implementation complexity due to their reliance on extensive human-annotated preference data or meticulously engineered quality dimensions that are often incomplete and engineering-intensive. Inspired by adversarial training in generative adversarial networks (GANs), this paper proposes GAN-RM, an efficient reward modeling framework that eliminates manual preference annotation and explicit quality dimension engineering. Our method trains the reward model through discrimination between a small set of representative, unpaired target samples(denoted as Preference Proxy Data) and model-generated ordinary outputs, requiring only a few hundred target samples. Comprehensive experiments demonstrate our GAN-RM's effectiveness across multiple key applications including test-time scaling implemented as Best-of-N sample filtering, post-training approaches like Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO). Code and data will be released at https://github.com/Visualignment/GAN-RM. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 17:59:40 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:39:32 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Liu",
"Runtao",
""
],
[
"Zhan",
"Jiahao",
""
],
[
"He",
"Yingqing",
""
],
[
"Wei",
"Chen",
""
],
[
"Yuille",
"Alan",
""
],
[
"Chen",
"Qifeng",
""
]
] |
2506.13857 | Roberto Tejada Arevalo | Roberto Tejada Arevalo | Different Inhomogeneous Evolutionary Histories for Uranus and Neptune | 17 pages, 6 figures, 3 tables, submitted to ApJL. This updated version removes Figure 6 to avoid density-total mass confusion coming from the density bounds of Movshovitz & Fortney (2022). I discuss these in the text instead | null | null | null | astro-ph.EP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present updated non-adiabatic and inhomogeneous evolution models for Uranus and Neptune, employing an interior composition of methane, ammonia, water, and rocks. Following formation trends of the gas giants, Uranus and Neptune formation models are applied, where both planets begin with layers stable to convection. Both planets are subject to convective mixing throughout their evolution. Uranus undergoes modest convective mixing, preserving much of its primordial internal heat. In contrast, Neptune's interior undergoes extensive mixing, homogenization, and adiabatic cooling of the outer 40\% of its envelope. The subsequent release of internal energy in Neptune, driven by the convective instability of its primordial outer compositional gradient, accounts for its higher luminosity relative to Uranus. Thus, the observed luminosity differences between Uranus and Neptune could be primarily dictated by the convective stability of their outer envelopes. The extensive convective mixing in Neptune leads to a higher metallicity in its outer region compared to Uranus, a feature seen in atmospheric measurements and shown in past interior models of Neptune. Due to Neptune's more pronounced cooling, our models predict favorable conditions for hydrogen-water immiscibility in its envelope. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 18:00:00 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:18:25 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Arevalo",
"Roberto Tejada",
""
]
] |
2506.13897 | Thomas Kreutz | Thomas Kreutz, Max M\"uhlh\"auser, Alejandro Sanchez Guinea | DeSPITE: Exploring Contrastive Deep Skeleton-Pointcloud-IMU-Text Embeddings for Advanced Point Cloud Human Activity Understanding | Accepted to ICCV 2025 | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Despite LiDAR (Light Detection and Ranging) being an effective privacy-preserving alternative to RGB cameras to perceive human activities, it remains largely underexplored in the context of multi-modal contrastive pre-training for human activity understanding (e.g., human activity recognition (HAR), retrieval, or person re-identification (RE-ID)). To close this gap, our work explores learning the correspondence between LiDAR point clouds, human skeleton poses, IMU data, and text in a joint embedding space. More specifically, we present DeSPITE, a Deep Skeleton-Pointcloud-IMU-Text Embedding model, which effectively learns a joint embedding space across these four modalities. At the heart of our empirical exploration, we have combined the existing LIPD and Babel datasets, which enabled us to synchronize data of all four modalities, allowing us to explore the learning of a new joint embedding space. Our experiments demonstrate novel human activity understanding tasks for point cloud sequences enabled through DeSPITE, including Skeleton<->Pointcloud<->IMU matching, retrieval, and temporal moment retrieval. Furthermore, we show that DeSPITE is an effective pre-training strategy for point cloud HAR through experiments in MSR-Action3D and HMPEAR. | [
{
"version": "v1",
"created": "Mon, 16 Jun 2025 18:18:44 GMT"
},
{
"version": "v2",
"created": "Fri, 20 Jun 2025 14:16:26 GMT"
},
{
"version": "v3",
"created": "Wed, 25 Jun 2025 19:20:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kreutz",
"Thomas",
""
],
[
"Mühlhäuser",
"Max",
""
],
[
"Guinea",
"Alejandro Sanchez",
""
]
] |
2506.14539 | Yeonghwan Shin | Daewon Kang, YeongHwan Shin, Doyeon Kim, Kyu-Hwan Jung, Meong Hi Son | Doppelganger Method: Breaking Role Consistency in LLM Agent via Prompt-based Transferable Adversarial Attack | null | null | null | null | cs.AI cs.CR | http://creativecommons.org/licenses/by/4.0/ | Since the advent of large language models, prompt engineering now enables the rapid, low-effort creation of diverse autonomous agents that are already in widespread use. Yet this convenience raises urgent concerns about the safety, robustness, and behavioral consistency of the underlying prompts, along with the pressing challenge of preventing those prompts from being exposed to user's attempts. In this paper, we propose the ''Doppelganger method'' to demonstrate the risk of an agent being hijacked, thereby exposing system instructions and internal information. Next, we define the ''Prompt Alignment Collapse under Adversarial Transfer (PACAT)'' level to evaluate the vulnerability to this adversarial transfer attack. We also propose a ''Caution for Adversarial Transfer (CAT)'' prompt to counter the Doppelganger method. The experimental results demonstrate that the Doppelganger method can compromise the agent's consistency and expose its internal information. In contrast, CAT prompts enable effective defense against this adversarial attack. | [
{
"version": "v1",
"created": "Tue, 17 Jun 2025 14:01:39 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 05:18:19 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kang",
"Daewon",
""
],
[
"Shin",
"YeongHwan",
""
],
[
"Kim",
"Doyeon",
""
],
[
"Jung",
"Kyu-Hwan",
""
],
[
"Son",
"Meong Hi",
""
]
] |
2506.14868 | Jo\~ao Sieiro Dos Santos | Jo\~ao S. Santos, Vitor Cardoso, Jos\'e Nat\'ario, Maarten van de Meent | Gravitational waves from b-EMRIs: Doppler shift and beaming, resonant excitation, helicity oscillations and self-lensing | 9+14 pages, 3+3 figures. Matches the version submitted to PRL | null | null | null | gr-qc astro-ph.HE physics.class-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study gravitational waves from a stellar-mass binary orbiting a spinning supermassive black hole, a system referred to as a binary extreme mass ratio inspiral (b-EMRI). We use Dixon's formalism to describe the stellar-mass binary as a particle with internal structure, and keep terms up to quadrupole order to capture the generation of gravitational waves by the inner motion of the stellar-mass binary. The problem of emission and propagation of waves is treated from first principles using black hole perturbation theory. In the gravitational waveform at future null infinity, we identify for the first time Doppler shifts and beaming due to the motion of the center of mass, as well as helicity breaking gravitational lensing, and resonances with ringdown modes of the supermassive black hole. We establish that previously proposed phenomenological models inadequately capture these effects. | [
{
"version": "v1",
"created": "Tue, 17 Jun 2025 18:00:00 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 18:00:00 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Santos",
"João S.",
""
],
[
"Cardoso",
"Vitor",
""
],
[
"Natário",
"José",
""
],
[
"van de Meent",
"Maarten",
""
]
] |
2506.15002 | Dorian Araya | Kyra Jacobo and Dorian Araya | Extended datasamples under the lens of Brane World Theory | 14 pages, 8 figures, 3 tables | Class.Quant.Grav. 42 (2025) 11, 115014 | 10.1088/1361-6382/adda89 | null | astro-ph.CO gr-qc | http://creativecommons.org/licenses/by/4.0/ | This work revises the Brane World Theory known as Randall-Sundrum with the modification of an exponential, redshift-dependent brane tension. This model is studied in a scenario assuming no dark energy, with the aim of determining whether it can reproduce the universe's acceleration on its own, without the addition of a dark energy fluid. Bayesian statistical analysis is performed in order to constrain the free parameters of each scenario using SLS, SNIa, OHD, and BAO data samples, the last two considering newly added data points. Both Planck and Riess priors for h are used and compared. In both cases, we are able to reproduce the late-time accelerated expansion in agreement with observational data. Interesting consistencies at the transition redshift zt with LCDM are found, suggesting that this might be a suitable model for studying the evolution of the universe up to the present date. However, some pathologies are detected in this model, namely a "Big Rip" divergence of H(z) at z = 1, as well as a strong dependence between the functional form of the brane tension and the future evolution of the universe. | [
{
"version": "v1",
"created": "Tue, 17 Jun 2025 22:17:05 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 22:35:34 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Jacobo",
"Kyra",
""
],
[
"Araya",
"Dorian",
""
]
] |
2506.15799 | Andrew Wagenmaker | Andrew Wagenmaker and Mitsuhiko Nakamoto and Yunchu Zhang and Seohong Park and Waleed Yagoub and Anusha Nagabandi and Abhishek Gupta and Sergey Levine | Steering Your Diffusion Policy with Latent Space Reinforcement Learning | null | null | null | null | cs.RO cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Robotic control policies learned from human demonstrations have achieved impressive results in many real-world applications. However, in scenarios where initial performance is not satisfactory, as is often the case in novel open-world settings, such behavioral cloning (BC)-learned policies typically require collecting additional human demonstrations to further improve their behavior -- an expensive and time-consuming process. In contrast, reinforcement learning (RL) holds the promise of enabling autonomous online policy improvement, but often falls short of achieving this due to the large number of samples it typically requires. In this work we take steps towards enabling fast autonomous adaptation of BC-trained policies via efficient real-world RL. Focusing in particular on diffusion policies -- a state-of-the-art BC methodology -- we propose diffusion steering via reinforcement learning (DSRL): adapting the BC policy by running RL over its latent-noise space. We show that DSRL is highly sample efficient, requires only black-box access to the BC policy, and enables effective real-world autonomous policy improvement. Furthermore, DSRL avoids many of the challenges associated with finetuning diffusion policies, obviating the need to modify the weights of the base policy at all. We demonstrate DSRL on simulated benchmarks, real-world robotic tasks, and for adapting pretrained generalist policies, illustrating its sample efficiency and effective performance at real-world policy improvement. | [
{
"version": "v1",
"created": "Wed, 18 Jun 2025 18:35:57 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 19:09:52 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wagenmaker",
"Andrew",
""
],
[
"Nakamoto",
"Mitsuhiko",
""
],
[
"Zhang",
"Yunchu",
""
],
[
"Park",
"Seohong",
""
],
[
"Yagoub",
"Waleed",
""
],
[
"Nagabandi",
"Anusha",
""
],
[
"Gupta",
"Abhishek",
""
],
[
"Levine",
"Sergey",
""
]
] |
2506.15830 | Riccardo Di Sipio | Riccardo Di Sipio | Rethinking LLM Training through Information Geometry and Quantum Metrics | 9 pages, 1 figure(s) | null | null | null | cs.CL quant-ph | http://creativecommons.org/licenses/by/4.0/ | Optimization in large language models (LLMs) unfolds over high-dimensional parameter spaces with non-Euclidean structure. Information geometry frames this landscape using the Fisher information metric, enabling more principled learning via natural gradient descent. Though often impractical, this geometric lens clarifies phenomena such as sharp minima, generalization, and observed scaling laws. We argue that curvature-aware approaches deepen our understanding of LLM training. Finally, we speculate on quantum analogies based on the Fubini-Study metric and Quantum Fisher Information, hinting at efficient optimization in quantum-enhanced systems. | [
{
"version": "v1",
"created": "Wed, 18 Jun 2025 19:17:47 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:14:42 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Di Sipio",
"Riccardo",
""
]
] |
2506.15895 | Vincent Guigues | Pablo Barros and Roger Behling and Vincent Guigues | Parallel Polyhedral Projection Method for the Convex Feasibility Problem | null | null | null | null | math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper, we introduce and study the Parallel Polyhedral Projection Method (3PM) and the Approximate Parallel Polyhedral Projection Method (A3PM) for finding a point in the intersection of finitely many closed convex sets. Each iteration has two phases: parallel projections onto the target sets (exact in 3PM, approximate in A3PM), followed by an exact or approximate projection onto a polyhedron defined by supporting half-spaces. These strategies appear novel, as existing methods largely focus on parallel schemes like Cimmino's method. Numerical experiments demonstrate that A3PM often outperforms both classical and recent projection-based methods when the number of sets is greater than two. Theoretically, we establish global convergence for both 3PM and A3PM without regularity assumptions. Under a Slater condition or error bound, we prove linear convergence, even with inexact projections. Additionally, we show that 3PM achieves superlinear convergence under suitable geometric assumptions. | [
{
"version": "v1",
"created": "Wed, 18 Jun 2025 21:39:35 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 13:42:42 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 14:05:01 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Barros",
"Pablo",
""
],
[
"Behling",
"Roger",
""
],
[
"Guigues",
"Vincent",
""
]
] |
2506.15928 | Myke C. Cohen | Myke C. Cohen, Zhe Su, Hsien-Te Kao, Daniel Nguyen, Spencer Lynch, Maarten Sap, Svitlana Volkova | Exploring Big Five Personality and AI Capability Effects in LLM-Simulated Negotiation Dialogues | Under review for KDD 2025 Workshop on Evaluation and Trustworthiness of Agentic and Generative AI Models | null | null | null | cs.AI cs.CL cs.HC | http://creativecommons.org/licenses/by-nc-sa/4.0/ | This paper presents an evaluation framework for agentic AI systems in mission-critical negotiation contexts, addressing the need for AI agents that can adapt to diverse human operators and stakeholders. Using Sotopia as a simulation testbed, we present two experiments that systematically evaluated how personality traits and AI agent characteristics influence LLM-simulated social negotiation outcomes--a capability essential for a variety of applications involving cross-team coordination and civil-military interactions. Experiment 1 employs causal discovery methods to measure how personality traits impact price bargaining negotiations, through which we found that Agreeableness and Extraversion significantly affect believability, goal achievement, and knowledge acquisition outcomes. Sociocognitive lexical measures extracted from team communications detected fine-grained differences in agents' empathic communication, moral foundations, and opinion patterns, providing actionable insights for agentic AI systems that must operate reliably in high-stakes operational scenarios. Experiment 2 evaluates human-AI job negotiations by manipulating both simulated human personality and AI system characteristics, specifically transparency, competence, adaptability, demonstrating how AI agent trustworthiness impact mission effectiveness. These findings establish a repeatable evaluation methodology for experimenting with AI agent reliability across diverse operator personalities and human-agent team dynamics, directly supporting operational requirements for reliable AI systems. Our work advances the evaluation of agentic AI workflows by moving beyond standard performance metrics to incorporate social dynamics essential for mission success in complex operations. | [
{
"version": "v1",
"created": "Thu, 19 Jun 2025 00:14:56 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 23:42:18 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Cohen",
"Myke C.",
""
],
[
"Su",
"Zhe",
""
],
[
"Kao",
"Hsien-Te",
""
],
[
"Nguyen",
"Daniel",
""
],
[
"Lynch",
"Spencer",
""
],
[
"Sap",
"Maarten",
""
],
[
"Volkova",
"Svitlana",
""
]
] |
2506.16182 | Vilhelm Lithell | Elias Jarlebring, Vilhelm P. Lithell | From eigenvector nonlinearities to eigenvalue nonlinearities | 19 pages, 5 figures | null | null | null | math.NA cs.NA | http://creativecommons.org/licenses/by/4.0/ | Over the past decades, transformations between different classes of eigenvalue problems have played a central role in the development of numerical methods for eigenvalue computations. One of the most well-known and successful examples of this is the companion linearization for polynomial eigenvalue problems. In this paper, we construct a transformation that equivalently reframes a specific type of eigenvalue problem with eigenvector nonlinearities (NEPv) into an eigenvalue problem with eigenvalue nonlinearities (NEP). The NEPv class considered consists of nonlinearities expressed as sums of products of matrices and scalar functions, where the scalar functions depend nonlinearly on the eigenvector. Our transformation defines scalar eigenvalue nonlinearities through a polynomial system, resulting in NEP nonlinearities of algebraic type. We propose methods to solve the polynomial system, one of which involves a multiparameter eigenvalue problem (MEP). We adapt well-established NEP solvers to this setting, with the most effective strategy being a combination of deflation and a locally quadratically convergent iterative method. The efficiency and properties of the approach are illustrated by solving a problem related to a modification of a Gross-Pitaevskii equation (GPE). The simulations are reproducible and publicly available. | [
{
"version": "v1",
"created": "Thu, 19 Jun 2025 10:02:27 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 12:29:55 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Jarlebring",
"Elias",
""
],
[
"Lithell",
"Vilhelm P.",
""
]
] |
2506.16398 | Peixiang Huang | Peixiang Huang, Yanyan Huang, Weiqin Zhao, Junjun He, Lequan Yu | HyperPath: Knowledge-Guided Hyperbolic Semantic Hierarchy Modeling for WSI Analysis | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Pathology is essential for cancer diagnosis, with multiple instance learning (MIL) widely used for whole slide image (WSI) analysis. WSIs exhibit a natural hierarchy -- patches, regions, and slides -- with distinct semantic associations. While some methods attempt to leverage this hierarchy for improved representation, they predominantly rely on Euclidean embeddings, which struggle to fully capture semantic hierarchies. To address this limitation, we propose HyperPath, a novel method that integrates knowledge from textual descriptions to guide the modeling of semantic hierarchies of WSIs in hyperbolic space, thereby enhancing WSI classification. Our approach adapts both visual and textual features extracted by pathology vision-language foundation models to the hyperbolic space. We design an Angular Modality Alignment Loss to ensure robust cross-modal alignment, while a Semantic Hierarchy Consistency Loss further refines feature hierarchies through entailment and contradiction relationships and thus enhance semantic coherence. The classification is performed with geodesic distance, which measures the similarity between entities in the hyperbolic semantic hierarchy. This eliminates the need for linear classifiers and enables a geometry-aware approach to WSI analysis. Extensive experiments show that our method achieves superior performance across tasks compared to existing methods, highlighting the potential of hyperbolic embeddings for WSI analysis. | [
{
"version": "v1",
"created": "Thu, 19 Jun 2025 15:30:33 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 09:10:30 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Huang",
"Peixiang",
""
],
[
"Huang",
"Yanyan",
""
],
[
"Zhao",
"Weiqin",
""
],
[
"He",
"Junjun",
""
],
[
"Yu",
"Lequan",
""
]
] |
2506.16438 | Shi-Mei Ma | Shi-Mei Ma, Jun-Ying Liu, Jean Yeh, Yeong-Nan Yeh | Eulerian-type polynomials over Stirling permutations and box sorting algorithm | 23 pages. arXiv admin note: substantial text overlap with arXiv:2312.02830 | null | null | null | math.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | It is well known that ascents, descents and plateaux are equidistributed over the set of classical Stirling permutations. Their common enumerative polynomials are the second-order Eulerian polynomials, which have been extensively studied by many researchers. This paper is divided into three parts. The first parts gives a convolution formula for the second-order Eulerian polynomials, which simplifies a result of Gessel. As an application, a determinantal expression for the second-order Eulerian polynomials is obtained. We then investigate the convolution formula of the trivariate second-order Eulerian polynomials. Among other things, by introducing three new statistics: proper ascent-plateau, improper ascent-plateau and trace, we discover that a six-variable Eulerian-type polynomial over a class of restricted Stirling permutations equals a six-variable Eulerian-type polynomial over signed permutations. By special parametrizations, we make use of Stirling permutations to give a unified interpretations of the $(p,q)$-Eulerian polynomials and derangement polynomials of types $A$ and $B$. The third part presents a box sorting algorithm which leads to a bijection between the terms in the expansion of $(cD)^nc$ and ordered weak set partitions, where $c$ is a smooth function in the indeterminate $x$ and $D$ is the derivative with respect to $x$. Using a map from ordered weak set partitions to standard Young tableaux, we find an expansion of $(cD)^nc$ in terms of standard Young tableaux. Combining this with grammars, we provide three interpretations of the second-order Eulerian polynomials. | [
{
"version": "v1",
"created": "Thu, 19 Jun 2025 16:15:40 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 02:25:49 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Ma",
"Shi-Mei",
""
],
[
"Liu",
"Jun-Ying",
""
],
[
"Yeh",
"Jean",
""
],
[
"Yeh",
"Yeong-Nan",
""
]
] |
2506.16486 | Gauranga Kumar Baishya Kumar Baishya | Gauranga Kumar Baishya | An introduction to Causal Modelling | null | null | null | null | stat.ME math.ST stat.ML stat.TH | http://creativecommons.org/licenses/by/4.0/ | This tutorial provides a concise introduction to modern causal modeling by integrating potential outcomes and graphical methods. We motivate causal questions such as counterfactual reasoning under interventions and define binary treatments and potential outcomes. We discuss causal effect measures-including average treatment effects on the treated and on the untreated-and choices of effect scales for binary outcomes. We derive identification in randomized experiments under exchangeability and consistency, and extend to stratification and blocking designs. We present inverse probability weighting with propensity score estimation and robust inference via sandwich estimators. Finally, we introduce causal graphs, d-separation, the backdoor criterion, single-world intervention graphs, and structural equation models, showing how graphical and potential-outcome approaches complement each other. Emphasis is placed on clear notation, intuitive explanations, and practical examples for applied researchers. | [
{
"version": "v1",
"created": "Thu, 19 Jun 2025 17:29:09 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:46:06 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Baishya",
"Gauranga Kumar",
""
]
] |
2506.16656 | Yaozhong Shi | Yaozhong Shi and Zachary E. Ross and Domniki Asimaki and Kamyar Azizzadenesheli | Mesh-Informed Neural Operator : A Transformer Generative Approach | null | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Generative models in function spaces, situated at the intersection of generative modeling and operator learning, are attracting increasing attention due to their immense potential in diverse scientific and engineering applications. While functional generative models are theoretically domain- and discretization-agnostic, current implementations heavily rely on the Fourier Neural Operator (FNO), limiting their applicability to regular grids and rectangular domains. To overcome these critical limitations, we introduce the Mesh-Informed Neural Operator (MINO). By leveraging graph neural operators and cross-attention mechanisms, MINO offers a principled, domain- and discretization-agnostic backbone for generative modeling in function spaces. This advancement significantly expands the scope of such models to more diverse applications in generative, inverse, and regression tasks. Furthermore, MINO provides a unified perspective on integrating neural operators with general advanced deep learning architectures. Finally, we introduce a suite of standardized evaluation metrics that enable objective comparison of functional generative models, addressing another critical gap in the field. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 00:00:22 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:45:03 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Shi",
"Yaozhong",
""
],
[
"Ross",
"Zachary E.",
""
],
[
"Asimaki",
"Domniki",
""
],
[
"Azizzadenesheli",
"Kamyar",
""
]
] |
2506.16663 | Michael Gyimadu | Michael Gyimadu, Gregory Bell, Ph.D | A Comparative Analysis of Principal Component Analysis (PCA) and Singular Value Decomposition (SVD) as Dimensionality Reduction Techniques | null | null | null | null | cs.CV cs.NA math.NA | http://creativecommons.org/licenses/by/4.0/ | High-dimensional image data often require dimensionality reduction before further analysis. This paper provides a purely analytical comparison of two linear techniques-Principal Component Analysis (PCA) and Singular Value Decomposition (SVD). After the derivation of each algorithm from first principles, we assess their interpretability, numerical stability, and suitability for differing matrix shapes. We synthesize rule-of-thumb guidelines for choosing one out of the two algorithms without empirical benchmarking, building on classical and recent numerical literature. Limitations and directions for future experimental work are outlined at the end. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 00:19:45 GMT"
},
{
"version": "v2",
"created": "Mon, 23 Jun 2025 01:58:54 GMT"
},
{
"version": "v3",
"created": "Wed, 25 Jun 2025 18:39:32 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Gyimadu",
"Michael",
""
],
[
"Bell",
"Gregory",
""
],
[
"D",
"Ph.",
""
]
] |
2506.16889 | Junghyun (Tony) Koo | Junghyun Koo, Marco A. Martinez-Ramirez, Wei-Hsiang Liao, Giorgio Fabbro, Michele Mancusi, and Yuki Mitsufuji | ITO-Master: Inference-Time Optimization for Audio Effects Modeling of Music Mastering Processors | ISMIR 2025 | null | null | null | cs.SD eess.AS | http://creativecommons.org/licenses/by/4.0/ | Music mastering style transfer aims to model and apply the mastering characteristics of a reference track to a target track, simulating the professional mastering process. However, existing methods apply fixed processing based on a reference track, limiting users' ability to fine-tune the results to match their artistic intent. In this paper, we introduce the ITO-Master framework, a reference-based mastering style transfer system that integrates Inference-Time Optimization (ITO) to enable finer user control over the mastering process. By optimizing the reference embedding during inference, our approach allows users to refine the output dynamically, making micro-level adjustments to achieve more precise mastering results. We explore both black-box and white-box methods for modeling mastering processors and demonstrate that ITO improves mastering performance across different styles. Through objective evaluation, subjective listening tests, and qualitative analysis using text-based conditioning with CLAP embeddings, we validate that ITO enhances mastering style similarity while offering increased adaptability. Our framework provides an effective and user-controllable solution for mastering style transfer, allowing users to refine their results beyond the initial style transfer. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 10:21:56 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 08:38:02 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Koo",
"Junghyun",
""
],
[
"Martinez-Ramirez",
"Marco A.",
""
],
[
"Liao",
"Wei-Hsiang",
""
],
[
"Fabbro",
"Giorgio",
""
],
[
"Mancusi",
"Michele",
""
],
[
"Mitsufuji",
"Yuki",
""
]
] |
2506.17053 | Jozef Genzor | Jozef Genzor, Roman Kr\v{c}m\'ar, Hiroshi Ueda, Denis Kochan, Andrej Gendiar, and Tomotoshi Nishino | Phase Transition of the Ising Model on a 3-Dimensional Fractal Lattice | 7 pages, 5 figures, 2 tables | null | null | null | cond-mat.stat-mech | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The critical behavior of the classical Ising model on a three-dimensional fractal lattice with Hausdorff dimension $d_H = \ln32 / \ln4 = 2.5$ is investigated using the higher-order tensor renormalization group (HOTRG) method. We determine the critical temperature $T_c \approx 2.65231$ and the critical exponents for magnetization $\beta \approx 0.059$ and field response $\delta \approx 35$. Unlike a previously studied 2D fractal with $d_H \approx 1.792$, the specific heat for this 3D fractal exhibits a divergent singularity at $T_c$. The results are compared with those for regular lattices and other fractal structures to elucidate the role of dimensionality in critical phenomena. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 15:04:37 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:19:56 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Genzor",
"Jozef",
""
],
[
"Krčmár",
"Roman",
""
],
[
"Ueda",
"Hiroshi",
""
],
[
"Kochan",
"Denis",
""
],
[
"Gendiar",
"Andrej",
""
],
[
"Nishino",
"Tomotoshi",
""
]
] |
2506.17450 | Jiacheng Chen | Jiacheng Chen, Ramin Mehran, Xuhui Jia, Saining Xie, Sanghyun Woo | BlenderFusion: 3D-Grounded Visual Editing and Generative Compositing | Project page: https://blenderfusion.github.io | null | null | null | cs.GR cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present BlenderFusion, a generative visual compositing framework that synthesizes new scenes by recomposing objects, camera, and background. It follows a layering-editing-compositing pipeline: (i) segmenting and converting visual inputs into editable 3D entities (layering), (ii) editing them in Blender with 3D-grounded control (editing), and (iii) fusing them into a coherent scene using a generative compositor (compositing). Our generative compositor extends a pre-trained diffusion model to process both the original (source) and edited (target) scenes in parallel. It is fine-tuned on video frames with two key training strategies: (i) source masking, enabling flexible modifications like background replacement; (ii) simulated object jittering, facilitating disentangled control over objects and camera. BlenderFusion significantly outperforms prior methods in complex compositional scene editing tasks. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 19:38:34 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 02:46:40 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Chen",
"Jiacheng",
""
],
[
"Mehran",
"Ramin",
""
],
[
"Jia",
"Xuhui",
""
],
[
"Xie",
"Saining",
""
],
[
"Woo",
"Sanghyun",
""
]
] |
2506.17501 | Shreeram Athreya | Shreeram Athreya, Carlos Olivares, Ameera Ismail, Kambiz Nael, William Speier, and Corey Arnold | DSA-NRP: No-Reflow Prediction from Angiographic Perfusion Dynamics in Stroke EVT | 12 pages, 4 figures | null | null | null | eess.IV cs.CV | http://creativecommons.org/licenses/by/4.0/ | Following successful large-vessel recanalization via endovascular thrombectomy (EVT) for acute ischemic stroke (AIS), some patients experience a complication known as no-reflow, defined by persistent microvascular hypoperfusion that undermines tissue recovery and worsens clinical outcomes. Although prompt identification is crucial, standard clinical practice relies on perfusion magnetic resonance imaging (MRI) within 24 hours post-procedure, delaying intervention. In this work, we introduce the first-ever machine learning (ML) framework to predict no-reflow immediately after EVT by leveraging previously unexplored intra-procedural digital subtraction angiography (DSA) sequences and clinical variables. Our retrospective analysis included AIS patients treated at UCLA Medical Center (2011-2024) who achieved favorable mTICI scores (2b-3) and underwent pre- and post-procedure MRI. No-reflow was defined as persistent hypoperfusion (Tmax > 6 s) on post-procedural imaging. From DSA sequences (AP and lateral views), we extracted statistical and temporal perfusion features from the target downstream territory to train ML classifiers for predicting no-reflow. Our novel method significantly outperformed a clinical-features baseline(AUC: 0.7703 $\pm$ 0.12 vs. 0.5728 $\pm$ 0.12; accuracy: 0.8125 $\pm$ 0.10 vs. 0.6331 $\pm$ 0.09), demonstrating that real-time DSA perfusion dynamics encode critical insights into microvascular integrity. This approach establishes a foundation for immediate, accurate no-reflow prediction, enabling clinicians to proactively manage high-risk patients without reliance on delayed imaging. | [
{
"version": "v1",
"created": "Fri, 20 Jun 2025 22:40:51 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 23:15:21 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Athreya",
"Shreeram",
""
],
[
"Olivares",
"Carlos",
""
],
[
"Ismail",
"Ameera",
""
],
[
"Nael",
"Kambiz",
""
],
[
"Speier",
"William",
""
],
[
"Arnold",
"Corey",
""
]
] |
2506.17572 | Xu Zhiqiang | Zhiqiang Xu | Signal Recovery on Algebraic Varieties Using Linear Samples | 16 pages | null | null | null | cs.IT math.AG math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The recovery of an unknown signal from its linear measurements is a fundamental problem spanning numerous scientific and engineering disciplines. Commonly, prior knowledge suggests that the underlying signal resides within a known algebraic variety. This context naturally leads to a question: what is the minimum number of measurements required to uniquely recover any signal belonging to such an algebraic variety? In this survey paper, we introduce a method that leverages tools from algebraic geometry to address this question. We then demonstrate the utility of this approach by applying it to two problems: phase retrieval and low-rank matrix recovery. We also highlight several open problems, which could serve as a basis for future investigations in this field. | [
{
"version": "v1",
"created": "Sat, 21 Jun 2025 03:56:16 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 05:42:53 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Xu",
"Zhiqiang",
""
]
] |
2506.17589 | Bowen Wang | Bowen Wang, Zhouqiang Jiang, Yasuaki Susumu, Shotaro Miwa, Tianwei Chen, Yuta Nakashima | Taming the Untamed: Graph-Based Knowledge Retrieval and Reasoning for MLLMs to Conquer the Unknown | Accepted by ICCV 2025 | null | null | null | cs.AI | http://creativecommons.org/licenses/by/4.0/ | The real value of knowledge lies not just in its accumulation, but in its potential to be harnessed effectively to conquer the unknown. Although recent multimodal large language models (MLLMs) exhibit impressing multimodal capabilities, they often fail in rarely encountered domain-specific tasks due to limited relevant knowledge. To explore this, we adopt visual game cognition as a testbed and select Monster Hunter: World as the target to construct a multimodal knowledge graph (MH-MMKG), which incorporates multi-modalities and intricate entity relations. We also design a series of challenging queries based on MH-MMKG to evaluate the models' ability for complex knowledge retrieval and reasoning. Furthermore, we propose a multi-agent retriever that enables a model to autonomously search relevant knowledge without additional training. Experimental results show that our approach significantly enhances the performance of MLLMs, providing a new perspective on multimodal knowledge-augmented reasoning and laying a solid foundation for future research. | [
{
"version": "v1",
"created": "Sat, 21 Jun 2025 05:01:02 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 01:13:31 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wang",
"Bowen",
""
],
[
"Jiang",
"Zhouqiang",
""
],
[
"Susumu",
"Yasuaki",
""
],
[
"Miwa",
"Shotaro",
""
],
[
"Chen",
"Tianwei",
""
],
[
"Nakashima",
"Yuta",
""
]
] |
2506.17851 | Jin Ai | Jin Ai and Richard S. Steinberg and Chao Guo and Filipi Nascimento Silva | Triadic Novelty: A Typology and Measurement Framework for Recognizing Novel Contributions in Science | 27 pages, 3 figures, 5 tables | null | null | null | physics.soc-ph cs.CY cs.SI stat.AP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Scientific progress depends on novel ideas, but current reward systems often fail to recognize them. Many existing metrics conflate novelty with popularity, privileging ideas that fit existing paradigms over those that challenge them. This study develops a theory-driven framework to better understand how different types of novelty emerge, take hold, and receive recognition. Drawing on network science and theories of discovery, we introduce a triadic typology: Pioneers, who introduce entirely new topics; Mavericks, who recombine distant concepts; and Vanguards, who reinforce weak but promising connections. We apply this typology to a dataset of 41,623 articles in the interdisciplinary field of philanthropy and nonprofit studies, linking novelty types to five-year citation counts using mixed-effects negative binomial regression. Results show that novelty is not uniformly rewarded. Pioneer efforts are foundational but often overlooked. Maverick novelty shows consistent citation benefits, particularly rewarded when it displaces prior focus. Vanguard novelty is more likely to gain recognition when it strengthens weakly connected topics, but its citation advantage diminishes as those reinforced nodes become more central. To enable fair comparison across time and domains, we introduce a simulated baseline model. These findings improve the evaluation of innovations, affecting science policy, funding, and institutional assessment practices. | [
{
"version": "v1",
"created": "Sat, 21 Jun 2025 23:09:04 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 18:24:05 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Ai",
"Jin",
""
],
[
"Steinberg",
"Richard S.",
""
],
[
"Guo",
"Chao",
""
],
[
"Silva",
"Filipi Nascimento",
""
]
] |
2506.17859 | Daniel Wurgaft | Daniel Wurgaft, Ekdeep Singh Lubana, Core Francisco Park, Hidenori Tanaka, Gautam Reddy, Noah D. Goodman | In-Context Learning Strategies Emerge Rationally | Preprint | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent work analyzing in-context learning (ICL) has identified a broad set of strategies that describe model behavior in different experimental conditions. We aim to unify these findings by asking why a model learns these disparate strategies in the first place. Specifically, we start with the observation that when trained to learn a mixture of tasks, as is popular in the literature, the strategies learned by a model for performing ICL can be captured by a family of Bayesian predictors: a memorizing predictor, which assumes a discrete prior on the set of seen tasks, and a generalizing predictor, where the prior matches the underlying task distribution. Adopting the normative lens of rational analysis, where a learner's behavior is explained as an optimal adaptation to data given computational constraints, we develop a hierarchical Bayesian framework that almost perfectly predicts Transformer next-token predictions throughout training -- without assuming access to its weights. Under this framework, pretraining is viewed as a process of updating the posterior probability of different strategies, and inference-time behavior as a posterior-weighted average over these strategies' predictions. Our framework draws on common assumptions about neural network learning dynamics, which make explicit a tradeoff between loss and complexity among candidate strategies: beyond how well it explains the data, a model's preference towards implementing a strategy is dictated by its complexity. This helps explain well-known ICL phenomena, while offering novel predictions: e.g., we show a superlinear trend in the timescale for transitioning from generalization to memorization as task diversity increases. Overall, our work advances an explanatory and predictive account of ICL grounded in tradeoffs between strategy loss and complexity. | [
{
"version": "v1",
"created": "Sat, 21 Jun 2025 23:49:08 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:54:57 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Wurgaft",
"Daniel",
""
],
[
"Lubana",
"Ekdeep Singh",
""
],
[
"Park",
"Core Francisco",
""
],
[
"Tanaka",
"Hidenori",
""
],
[
"Reddy",
"Gautam",
""
],
[
"Goodman",
"Noah D.",
""
]
] |
2506.17915 | Cheng Zeng | Cheng Zeng, Gengji Li | Some sharp bounds on the average Steiner (k, l)-eccentricity for trees | 14 pages, 6 figures | null | null | null | math.CO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper we introduce some transformations for trees that do not increase the average Steiner $(k,l)$-eccentricity for all $0\leq l\leq k\leq n$. Using these transformations, we obtain some sharp bounds on the average Steiner $(k,l)$-eccentricity for trees with some certain conditions, including given nodes, given diameter, given max degree and given leaves, and get the corresponding extremal trees as well. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 06:34:59 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 12:06:05 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 13:28:12 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Zeng",
"Cheng",
""
],
[
"Li",
"Gengji",
""
]
] |
2506.17923 | Markus Weiger | Markus Weiger, Johan Overweg, Franciszek Hennel, Emily Louise Baadsvik, Samuel Bianchi, Oskar Bj\"orkqvist, Roger Luechinger, Jens Metzger, Eric Michael, Thomas Schmid, Lauro Singenberger, Urs Sturzenegger, Erik Oskam, Gerrit Vissers, Jos Koonen, Wout Schuth, Jeroen Koeleman, Martino Borgo Klaas Paul Pruessmann | A unipolar head gradient for high-field MRI without encoding ambiguity | null | null | null | null | physics.med-ph | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Purpose: MRI gradients with a conventional, bipolar design generally face a trade-off between performance, encoding ambiguity, and circumventing the latter by means of RF selectivity. This problem is particularly limiting in cutting-edge brain imaging performed at field strengths >= 7T and using high-performance head gradients. Methods: To address this issue, the present work proposes to fundamentally eliminate the encoding ambiguity in head gradients by using a unipolar z-gradient design that takes advantage of the signal-free range on one side of the imaging volume. This concept is demonstrated by implementation of a unipolar head gradient for operation at 7T. Results: Imaging in phantoms and in vivo demonstrates elimination of backfolding due to encoding ambiguity. At the same time, the unipolar design achieves efficiency on par with conventional bipolar design, resulting in high amplitude and slew-rate performance. Conclusion: The prospect of gradient systems based on a unipolar design holds promise for all advanced neuroimaging that demands high gradient performance. It will make the greatest difference at 7T and beyond where the absence of ambiguity removes a key concern and constraint in terms of RF behaviour and instrumentation. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 07:28:15 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:52:20 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Weiger",
"Markus",
""
],
[
"Overweg",
"Johan",
""
],
[
"Hennel",
"Franciszek",
""
],
[
"Baadsvik",
"Emily Louise",
""
],
[
"Bianchi",
"Samuel",
""
],
[
"Björkqvist",
"Oskar",
""
],
[
"Luechinger",
"Roger",
""
],
[
"Metzger",
"Jens",
""
],
[
"Michael",
"Eric",
""
],
[
"Schmid",
"Thomas",
""
],
[
"Singenberger",
"Lauro",
""
],
[
"Sturzenegger",
"Urs",
""
],
[
"Oskam",
"Erik",
""
],
[
"Vissers",
"Gerrit",
""
],
[
"Koonen",
"Jos",
""
],
[
"Schuth",
"Wout",
""
],
[
"Koeleman",
"Jeroen",
""
],
[
"Pruessmann",
"Martino Borgo Klaas Paul",
""
]
] |
2506.17955 | Ron Demjaha | Ron Demjaha and Konstantin Zarembo | String Integrability on the Coulomb Branch | null | null | null | null | hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The Coulomb branch of the N = 4 super-Yang-Mills theory is described by a D3-brane in the bulk of AdS_5 x S^5. We show that boundary conditions on this D3-brane preserve integrability of the AdS string. For that matter we construct dynamical and spectral parameter dependent reflection matrix that encodes an infinite number of conserved charges on the worldsheet of an open string ending on the D-brane. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 09:13:12 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:25:28 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Demjaha",
"Ron",
""
],
[
"Zarembo",
"Konstantin",
""
]
] |
2506.18019 | Yuanchen Bei | Yuanchen Bei, Weizhi Zhang, Siwen Wang, Weizhi Chen, Sheng Zhou, Hao Chen, Yong Li, Jiajun Bu, Shirui Pan, Yizhou Yu, Irwin King, Fakhri Karray, Philip S. Yu | Graphs Meet AI Agents: Taxonomy, Progress, and Future Opportunities | 20 pages, 7 figures | null | null | null | cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | AI agents have experienced a paradigm shift, from early dominance by reinforcement learning (RL) to the rise of agents powered by large language models (LLMs), and now further advancing towards a synergistic fusion of RL and LLM capabilities. This progression has endowed AI agents with increasingly strong abilities. Despite these advances, to accomplish complex real-world tasks, agents are required to plan and execute effectively, maintain reliable memory, and coordinate smoothly with other agents. Achieving these capabilities involves contending with ever-present intricate information, operations, and interactions. In light of this challenge, data structurization can play a promising role by transforming intricate and disorganized data into well-structured forms that agents can more effectively understand and process. In this context, graphs, with their natural advantage in organizing, managing, and harnessing intricate data relationships, present a powerful data paradigm for structurization to support the capabilities demanded by advanced AI agents. To this end, this survey presents a first systematic review of how graphs can empower AI agents. Specifically, we explore the integration of graph techniques with core agent functionalities, highlight notable applications, and identify prospective avenues for future research. By comprehensively surveying this burgeoning intersection, we hope to inspire the development of next-generation AI agents equipped to tackle increasingly sophisticated challenges with graphs. Related resources are collected and continuously updated for the community in the Github link. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 12:59:12 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:54:14 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Bei",
"Yuanchen",
""
],
[
"Zhang",
"Weizhi",
""
],
[
"Wang",
"Siwen",
""
],
[
"Chen",
"Weizhi",
""
],
[
"Zhou",
"Sheng",
""
],
[
"Chen",
"Hao",
""
],
[
"Li",
"Yong",
""
],
[
"Bu",
"Jiajun",
""
],
[
"Pan",
"Shirui",
""
],
[
"Yu",
"Yizhou",
""
],
[
"King",
"Irwin",
""
],
[
"Karray",
"Fakhri",
""
],
[
"Yu",
"Philip S.",
""
]
] |
2506.18028 | Junjian Li | Junjian Li, Hulin Kuang, Jin Liu, Hailin Yue, Mengshen He, and Jianxin Wang | MiCo: Multiple Instance Learning with Context-Aware Clustering for Whole Slide Image Analysis | MICCAI 2025 | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Multiple instance learning (MIL) has shown significant promise in histopathology whole slide image (WSI) analysis for cancer diagnosis and prognosis. However, the inherent spatial heterogeneity of WSIs presents critical challenges, as morphologically similar tissue types are often dispersed across distant anatomical regions. Conventional MIL methods struggle to model these scattered tissue distributions and capture cross-regional spatial interactions effectively. To address these limitations, we propose a novel Multiple instance learning framework with Context-Aware Clustering (MiCo), designed to enhance cross-regional intra-tissue correlations and strengthen inter-tissue semantic associations in WSIs. MiCo begins by clustering instances to distill discriminative morphological patterns, with cluster centroids serving as semantic anchors. To enhance cross-regional intra-tissue correlations, MiCo employs a Cluster Route module, which dynamically links instances of the same tissue type across distant regions via feature similarity. These semantic anchors act as contextual hubs, propagating semantic relationships to refine instance-level representations. To eliminate semantic fragmentation and strengthen inter-tissue semantic associations, MiCo integrates a Cluster Reducer module, which consolidates redundant anchors while enhancing information exchange between distinct semantic groups. Extensive experiments on two challenging tasks across nine large-scale public cancer datasets demonstrate the effectiveness of MiCo, showcasing its superiority over state-of-the-art methods. The code is available at https://github.com/junjianli106/MiCo. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 13:14:41 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 01:29:36 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Li",
"Junjian",
""
],
[
"Kuang",
"Hulin",
""
],
[
"Liu",
"Jin",
""
],
[
"Yue",
"Hailin",
""
],
[
"He",
"Mengshen",
""
],
[
"Wang",
"Jianxin",
""
]
] |
2506.18177 | MIngchao Liang | Mingchao Liang and Florian Meyer | Coherent Track-Before-Detect | null | null | null | null | eess.SP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Accurately tracking an unknown and time-varying number of objects in complex environments is a significant challenge but a fundamental capability in a variety of applications, including applied ocean sciences, surveillance, autonomous driving, and wireless communications. Conventional Bayesian multiobject tracking (MOT) methods typically employ a detect-then-track (DTT) approach, where a frontend detector preprocesses raw sensor data to extract measurements for MOT. The irreversible nature of this preprocessing step can discard valuable object-related information, particularly impairing the ability to resolve weak or closely spaced objects. The track-before-detect (TBD) paradigm offers an alternative by operating directly on sensor data. However, existing TBD approaches introduce simplifications to facilitate the development of inference methods, such as assuming known signal amplitudes or conditional independence between sensor measurements given object states. These assumptions can lead to suboptimal performance and limit the applicability of the resulting TBD methods in realistic scenarios.
This paper introduces coherent TBD based on a comprehensive signal model for sensor data. The new model accounts for sensor data correlations and amplitude fluctuations, enabling the accurate representation of the physics of the data-generating process in TBD. Coherent TBD is suitable for a wide range of problems in active and passive radar, active and passive sonar, as well as integrated sensing and communication systems. Based on a factor graph representation of the new measurement model, a scalable belief propagation (BP) method is developed to perform efficient Bayesian inference. Experimental results, performed with both synthetic and real data, demonstrate that the proposed method outperforms state-of-the-art conventional MOT methods. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 21:20:56 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:57:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Liang",
"Mingchao",
""
],
[
"Meyer",
"Florian",
""
]
] |
2506.18221 | Xingyu Yang | Xingyu Alice Yang, Jianyu Zhang, L\'eon Bottou | These Are Not All the Features You Are Looking For: A Fundamental Bottleneck in Supervised Pretraining | 10 pages, 7 figures, Preprint. Under review | null | null | null | cs.LG cs.AI stat.ML | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Transfer learning is a cornerstone of modern machine learning, promising a way to adapt models pretrained on a broad mix of data to new tasks with minimal new data. However, a significant challenge remains in ensuring that transferred features are sufficient to handle unseen datasets, amplified by the difficulty of quantifying whether two tasks are "related". To address these challenges, we evaluate model transfer from a pretraining mixture to each of its component tasks, assessing whether pretrained features can match the performance of task-specific direct training. We identify a fundamental limitation in deep learning models -- an "information saturation bottleneck" -- where networks fail to learn new features once they encode similar competing features during training. When restricted to learning only a subset of key features during pretraining, models will permanently lose critical features for transfer and perform inconsistently on data distributions, even components of the training mixture. Empirical evidence from published studies suggests that this phenomenon is pervasive in deep learning architectures -- factors such as data distribution or ordering affect the features that current representation learning methods can learn over time. This study suggests that relying solely on large-scale networks may not be as effective as focusing on task-specific training, when available. We propose richer feature representations as a potential solution to better generalize across new datasets and, specifically, present existing methods alongside a novel approach, the initial steps towards addressing this challenge. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 01:04:29 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 13:50:38 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Yang",
"Xingyu Alice",
""
],
[
"Zhang",
"Jianyu",
""
],
[
"Bottou",
"Léon",
""
]
] |
2506.18246 | Xiangzhao Hao | Xiangzhao Hao, Kuan Zhu, Hongyu Guo, Haiyun Guo, Ning Jiang, Quan Lu, Ming Tang, Jinqiao Wang | Referring Expression Instance Retrieval and A Strong End-to-End Baseline | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Using natural language to query visual information is a fundamental need in real-world applications. Text-Image Retrieval (TIR) retrieves a target image from a gallery based on an image-level description, while Referring Expression Comprehension (REC) localizes a target object within a given image using an instance-level description. However, real-world applications often present more complex demands. Users typically query an instance-level description across a large gallery and expect to receive both relevant image and the corresponding instance location. In such scenarios, TIR struggles with fine-grained descriptions and object-level localization, while REC is limited in its ability to efficiently search large galleries and lacks an effective ranking mechanism. In this paper, we introduce a new task called \textbf{Referring Expression Instance Retrieval (REIR)}, which supports both instance-level retrieval and localization based on fine-grained referring expressions. First, we propose a large-scale benchmark for REIR, named REIRCOCO, constructed by prompting advanced vision-language models to generate high-quality referring expressions for instances in the MSCOCO and RefCOCO datasets. Second, we present a baseline method, Contrastive Language-Instance Alignment with Relation Experts (CLARE), which employs a dual-stream architecture to address REIR in an end-to-end manner. Given a referring expression, the textual branch encodes it into a query embedding. The visual branch detects candidate objects and extracts their instance-level visual features. The most similar candidate to the query is selected for bounding box prediction. CLARE is first trained on object detection and REC datasets to establish initial grounding capabilities, then optimized via Contrastive Language-Instance Alignment (CLIA) for improved retrieval across images. We will release our code and benchmark publicly. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 02:28:44 GMT"
},
{
"version": "v2",
"created": "Tue, 24 Jun 2025 03:38:39 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 11:31:17 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Hao",
"Xiangzhao",
""
],
[
"Zhu",
"Kuan",
""
],
[
"Guo",
"Hongyu",
""
],
[
"Guo",
"Haiyun",
""
],
[
"Jiang",
"Ning",
""
],
[
"Lu",
"Quan",
""
],
[
"Tang",
"Ming",
""
],
[
"Wang",
"Jinqiao",
""
]
] |
2506.18324 | Yufan Chen | Shiping Fu, Yufan Chen, Zhe Zhang, Xiaolan Qiu, Qixiang Ye | ARSAR-Net: Intelligent SAR Imaging with Adaptive Regularization | null | null | null | null | eess.SP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Deep unfolding networks have recently emerged as a promising approach for synthetic aperture radar (SAR) imaging. However, baseline unfolding networks, typically derived from iterative reconstruction algorithms such as the alternating direction method of multipliers (ADMM), lack generalization capability across scenes, primarily because their regularizers are empirically designed rather than learned from data. In this study, we introduce a learnable regularizer into the unfolding network and propose a SAR imaging network with adaptive regularization (ARSAR-Net), which aims to generalize across heterogeneous scenes including offshore ships, islands, urban areas, and mountainous terrain. Furthermore, two variants of ARSAR-Net are developed, targeting improved imaging efficiency and reconstruction quality, respectively. Extensive validation through simulated and real-data experiments demonstrates three key advantages of ARSAR-Net: (1) a 50% increase in imaging speed over existing unfolding networks, (2) a PSNR gain of up to 2.0 dB in imaging quality, and (3) enhanced adaptability to complex scenes. These advancements establish a new paradigm for computationally efficient and generalizable SAR imaging systems. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 06:16:08 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 10:31:38 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Fu",
"Shiping",
""
],
[
"Chen",
"Yufan",
""
],
[
"Zhang",
"Zhe",
""
],
[
"Qiu",
"Xiaolan",
""
],
[
"Ye",
"Qixiang",
""
]
] |
2506.18575 | Kaifeng Sheng | Kaifeng Sheng, Zheng Zhou, Yingliang Peng, Qianwei Wang | 2D Triangle Splatting for Direct Differentiable Mesh Training | 13 pages, 8 figures | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Differentiable rendering with 3D Gaussian primitives has emerged as a powerful method for reconstructing high-fidelity 3D scenes from multi-view images. While it offers improvements over NeRF-based methods, this representation still encounters challenges with rendering speed and advanced rendering effects, such as relighting and shadow rendering, compared to mesh-based models. In this paper, we propose 2D Triangle Splatting (2DTS), a novel method that replaces 3D Gaussian primitives with 2D triangle facelets. This representation naturally forms a discrete mesh-like structure while retaining the benefits of continuous volumetric modeling. By incorporating a compactness parameter into the triangle primitives, we enable direct training of photorealistic meshes. Our experimental results demonstrate that our triangle-based method, in its vanilla version (without compactness tuning), achieves higher fidelity compared to state-of-the-art Gaussian-based methods. Furthermore, our approach produces reconstructed meshes with superior visual quality compared to existing mesh reconstruction methods. Please visit our project page at https://gaoderender.github.io/triangle-splatting. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 12:26:47 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 06:46:05 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Sheng",
"Kaifeng",
""
],
[
"Zhou",
"Zheng",
""
],
[
"Peng",
"Yingliang",
""
],
[
"Wang",
"Qianwei",
""
]
] |
2506.18671 | Dai Yuqin | Yuqin Dai, Wanlu Zhu, Ronghui Li, Xiu Li, Zhenyu Zhang, Jun Li, Jian Yang | TCDiff++: An End-to-end Trajectory-Controllable Diffusion Model for Harmonious Music-Driven Group Choreography | null | null | null | null | cs.SD cs.CV cs.GR eess.AS | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Music-driven dance generation has garnered significant attention due to its wide range of industrial applications, particularly in the creation of group choreography. During the group dance generation process, however, most existing methods still face three primary issues: multi-dancer collisions, single-dancer foot sliding and abrupt swapping in the generation of long group dance. In this paper, we propose TCDiff++, a music-driven end-to-end framework designed to generate harmonious group dance. Specifically, to mitigate multi-dancer collisions, we utilize a dancer positioning embedding to better maintain the relative positioning among dancers. Additionally, we incorporate a distance-consistency loss to ensure that inter-dancer distances remain within plausible ranges. To address the issue of single-dancer foot sliding, we introduce a swap mode embedding to indicate dancer swapping patterns and design a Footwork Adaptor to refine raw motion, thereby minimizing foot sliding. For long group dance generation, we present a long group diffusion sampling strategy that reduces abrupt position shifts by injecting positional information into the noisy input. Furthermore, we integrate a Sequence Decoder layer to enhance the model's ability to selectively process long sequences. Extensive experiments demonstrate that our TCDiff++ achieves state-of-the-art performance, particularly in long-duration scenarios, ensuring high-quality and coherent group dance generation. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 14:15:20 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 14:19:44 GMT"
},
{
"version": "v3",
"created": "Thu, 26 Jun 2025 13:53:17 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Dai",
"Yuqin",
""
],
[
"Zhu",
"Wanlu",
""
],
[
"Li",
"Ronghui",
""
],
[
"Li",
"Xiu",
""
],
[
"Zhang",
"Zhenyu",
""
],
[
"Li",
"Jun",
""
],
[
"Yang",
"Jian",
""
]
] |
2506.18728 | Steven Kolawole | Steven Kolawole, Keshav Santhanam, Virginia Smith, and Pratiksha Thaker | PARALLELPROMPT: Extracting Parallelism from Large Language Model Queries | In Adaptive Foundation Models: Evolving AI for Personalized and Efficient Learning | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | LLM serving systems typically treat user prompts as monolithic inputs, optimizing inference through decoding tricks or inter-query batching. However, many real-world prompts contain latent semantic parallelism--decomposable structures where subtasks can be executed independently to reduce latency while preserving meaning. We introduce PARALLELPROMPT, the first benchmark for measuring intra-query parallelism in natural user prompts. Our dataset comprises over 37,000 real-world prompts from public LLM chat logs, each annotated with a structured schema capturing task templates, shared context, and iteration inputs. These schemas are extracted using LLM-assisted prompting with rule-based multilingual validation. To evaluate the benefits of decomposition, we provide an execution suite that benchmarks serial vs. parallel strategies, measuring latency, structural adherence, and semantic fidelity. Our results show that intra-query parallelism can be successfully parsed in over 75% of curated datasets, unlocking up to 5x speedups on tasks like translation, comprehension, and comparative analysis, with minimal quality degradation. By releasing this benchmark, curation pipeline, and evaluation suite, we provide the first standardized testbed for studying structure-aware execution in LLM serving pipelines. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 15:05:54 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 16:35:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kolawole",
"Steven",
""
],
[
"Santhanam",
"Keshav",
""
],
[
"Smith",
"Virginia",
""
],
[
"Thaker",
"Pratiksha",
""
]
] |
2506.18861 | Alexandru Chirv\u{a}situ L. | Alexandru Chirvasitu | Pointwise-relatively-compact subgroups and trivial-weight-free representations | v2 corrects a misapplication of a cited reference; 6 pages + references | null | null | null | math.GR math.GN math.RT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | A pointwise-elliptic subset of a topological group is one whose elements all generate relatively-compact subgroups. A connected locally compact group has a dense pointwise-elliptic subgroup if and only if it is an extension by a compact normal subgroup of a semidirect product $\mathbb{L}\rtimes \mathbb{K}$ with connected, simply-connected Lie $\mathbb{L}$, compact Lie $\mathbb{K}$, with the commutator subgroup $\mathbb{K}'$ acting on the Lie algebra $Lie(\mathbb{L})$ with no trivial weights. This extends and recovers a result of Kabenyuk's, providing the analogous classification with $\mathbb{G}$ assumed Lie connected, topologically perfect, with no non-trivial central elliptic elements. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 17:24:09 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 14:53:31 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Chirvasitu",
"Alexandru",
""
]
] |
2506.18927 | Shriyank Somvanshi | Shriyank Somvanshi, Md Monzurul Islam, Gaurab Chhetri, Rohit Chakraborty, Mahmuda Sultana Mimi, Sawgat Ahmed Shuvo, Kazi Sifatul Islam, Syed Aaqib Javed, Sharif Ahmed Rafat, Anandi Dutta, Subasish Das | From Tiny Machine Learning to Tiny Deep Learning: A Survey | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The rapid growth of edge devices has driven the demand for deploying artificial intelligence (AI) at the edge, giving rise to Tiny Machine Learning (TinyML) and its evolving counterpart, Tiny Deep Learning (TinyDL). While TinyML initially focused on enabling simple inference tasks on microcontrollers, the emergence of TinyDL marks a paradigm shift toward deploying deep learning models on severely resource-constrained hardware. This survey presents a comprehensive overview of the transition from TinyML to TinyDL, encompassing architectural innovations, hardware platforms, model optimization techniques, and software toolchains. We analyze state-of-the-art methods in quantization, pruning, and neural architecture search (NAS), and examine hardware trends from MCUs to dedicated neural accelerators. Furthermore, we categorize software deployment frameworks, compilers, and AutoML tools enabling practical on-device learning. Applications across domains such as computer vision, audio recognition, healthcare, and industrial monitoring are reviewed to illustrate the real-world impact of TinyDL. Finally, we identify emerging directions including neuromorphic computing, federated TinyDL, edge-native foundation models, and domain-specific co-design approaches. This survey aims to serve as a foundational resource for researchers and practitioners, offering a holistic view of the ecosystem and laying the groundwork for future advancements in edge AI. | [
{
"version": "v1",
"created": "Sat, 21 Jun 2025 05:09:07 GMT"
},
{
"version": "v2",
"created": "Wed, 25 Jun 2025 21:42:13 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Somvanshi",
"Shriyank",
""
],
[
"Islam",
"Md Monzurul",
""
],
[
"Chhetri",
"Gaurab",
""
],
[
"Chakraborty",
"Rohit",
""
],
[
"Mimi",
"Mahmuda Sultana",
""
],
[
"Shuvo",
"Sawgat Ahmed",
""
],
[
"Islam",
"Kazi Sifatul",
""
],
[
"Javed",
"Syed Aaqib",
""
],
[
"Rafat",
"Sharif Ahmed",
""
],
[
"Dutta",
"Anandi",
""
],
[
"Das",
"Subasish",
""
]
] |
2506.18933 | Sebastian Fuchs | Sebastian Fuchs | From Trial Division to Smooth Prime Indicators | 15 pages, 7 figures | null | null | null | math.NT | http://creativecommons.org/licenses/by/4.0/ | This work introduces a flexible framework for constructing smooth (C-infinity) analogues of classical arithmetic functions, based on the smoothing of a trigonometric representation of trial division. A foundational function P: R -> R of class C^1 is first presented, whose zeros for x > 2 correspond precisely to the odd primes. The limitations of this function motivate its generalization into a versatile framework. This framework is then applied to construct two novel functions: a smooth analogue of the divisor-counting function, P_tau, and a smooth analogue of the sum-of-divisors function, P_sigma. Both functions are proven to be of class C-infinity. It is shown that these new constructions possess a complete prime-zero property for all integers x >= 2. The robustness of this property for real numbers is analyzed for each function. It is demonstrated that P_tau provides a robust prime indicator for all real numbers, while the properties of P_sigma in this regard lead to an open question concerning the existence of non-integer zeros. The construction methodology, properties, and potential of these functions are discussed in detail. | [
{
"version": "v1",
"created": "Sun, 22 Jun 2025 00:39:25 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:00:54 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Fuchs",
"Sebastian",
""
]
] |
2506.18959 | Weizhi Zhang | Weizhi Zhang, Yangning Li, Yuanchen Bei, Junyu Luo, Guancheng Wan, Liangwei Yang, Chenxuan Xie, Yuyao Yang, Wei-Chieh Huang, Chunyu Miao, Henry Peng Zou, Xiao Luo, Yusheng Zhao, Yankai Chen, Chunkit Chan, Peilin Zhou, Xinyang Zhang, Chenwei Zhang, Jingbo Shang, Ming Zhang, Yangqiu Song, Irwin King, Philip S. Yu | From Web Search towards Agentic Deep Research: Incentivizing Search with Reasoning Agents | null | null | null | null | cs.IR cs.CL cs.LG | http://creativecommons.org/licenses/by/4.0/ | Information retrieval is a cornerstone of modern knowledge acquisition, enabling billions of queries each day across diverse domains. However, traditional keyword-based search engines are increasingly inadequate for handling complex, multi-step information needs. Our position is that Large Language Models (LLMs), endowed with reasoning and agentic capabilities, are ushering in a new paradigm termed Agentic Deep Research. These systems transcend conventional information search techniques by tightly integrating autonomous reasoning, iterative retrieval, and information synthesis into a dynamic feedback loop. We trace the evolution from static web search to interactive, agent-based systems that plan, explore, and learn. We also introduce a test-time scaling law to formalize the impact of computational depth on reasoning and search. Supported by benchmark results and the rise of open-source implementations, we demonstrate that Agentic Deep Research not only significantly outperforms existing approaches, but is also poised to become the dominant paradigm for future information seeking. All the related resources, including industry products, research papers, benchmark datasets, and open-source implementations, are collected for the community in https://github.com/DavidZWZ/Awesome-Deep-Research. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 17:27:19 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:18:00 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Zhang",
"Weizhi",
""
],
[
"Li",
"Yangning",
""
],
[
"Bei",
"Yuanchen",
""
],
[
"Luo",
"Junyu",
""
],
[
"Wan",
"Guancheng",
""
],
[
"Yang",
"Liangwei",
""
],
[
"Xie",
"Chenxuan",
""
],
[
"Yang",
"Yuyao",
""
],
[
"Huang",
"Wei-Chieh",
""
],
[
"Miao",
"Chunyu",
""
],
[
"Zou",
"Henry Peng",
""
],
[
"Luo",
"Xiao",
""
],
[
"Zhao",
"Yusheng",
""
],
[
"Chen",
"Yankai",
""
],
[
"Chan",
"Chunkit",
""
],
[
"Zhou",
"Peilin",
""
],
[
"Zhang",
"Xinyang",
""
],
[
"Zhang",
"Chenwei",
""
],
[
"Shang",
"Jingbo",
""
],
[
"Zhang",
"Ming",
""
],
[
"Song",
"Yangqiu",
""
],
[
"King",
"Irwin",
""
],
[
"Yu",
"Philip S.",
""
]
] |
2506.19014 | Abhay Kumar | Abhay Kumar, Kunal Verma, Omkar More | IndieFake Dataset: A Benchmark Dataset for Audio Deepfake Detection | Project Website: https://indie-fake-dataset.netlify.app/ | null | null | null | cs.SD cs.AI eess.AS | http://creativecommons.org/licenses/by/4.0/ | Advancements in audio deepfake technology offers benefits like AI assistants, better accessibility for speech impairments, and enhanced entertainment. However, it also poses significant risks to security, privacy, and trust in digital communications. Detecting and mitigating these threats requires comprehensive datasets. Existing datasets lack diverse ethnic accents, making them inadequate for many real-world scenarios. Consequently, models trained on these datasets struggle to detect audio deepfakes in diverse linguistic and cultural contexts such as in South-Asian countries. Ironically, there is a stark lack of South-Asian speaker samples in the existing datasets despite constituting a quarter of the worlds population. This work introduces the IndieFake Dataset (IFD), featuring 27.17 hours of bonafide and deepfake audio from 50 English speaking Indian speakers. IFD offers balanced data distribution and includes speaker-level characterization, absent in datasets like ASVspoof21 (DF). We evaluated various baselines on IFD against existing ASVspoof21 (DF) and In-The-Wild (ITW) datasets. IFD outperforms ASVspoof21 (DF) and proves to be more challenging compared to benchmark ITW dataset. The complete dataset, along with documentation and sample reference clips, is publicly accessible for research use on project website. | [
{
"version": "v1",
"created": "Mon, 23 Jun 2025 18:10:06 GMT"
},
{
"version": "v2",
"created": "Thu, 26 Jun 2025 17:21:45 GMT"
}
] | 2025-06-27T00:00:00 | [
[
"Kumar",
"Abhay",
""
],
[
"Verma",
"Kunal",
""
],
[
"More",
"Omkar",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.