Learn What's Left, Not What's Mastered: Saturation Aware Advantage Reweighting for Multi-Reward Policy Optimization
Paper • 2608.16072 • Published • 149
AI & ML interests
None defined yet.
Recent Activity
View all activity
williamium
authored 5
papers 6 days ago
Self-Supervised Visual On-Policy Distillation
Paper • 2608.14144 • Published • 170
On-Policy Self-Distillation without Any Supervision
Paper • 2608.06296 • Published • 218
Attention Sinks and Outliers in Attention Residuals
Paper • 2605.17887 • Published
Demystifying Agent Skills: Why They Work-Until They Don't
Paper • 2608.14036 • Published • 166
williamium
submitted a
paper to Daily Papers 7 days ago
williamium
submitted a
paper to Daily Papers 8 days ago
williamium
submitted a
paper to Daily Papers 9 days ago
williamium
submitted a
paper to Daily Papers 10 days ago
williamium
submitted a
paper to Daily Papers 16 days ago
williamium
authored 10
papers about 1 month ago
Visual Contrastive Self-Distillation
Paper • 2607.21556 • Published • 52
From RGB Generation to Dense Field Readout: Pixel-Space Dense Prediction with Text-to-Image Models
Paper • 2607.06553 • Published • 20
SPACENUM: Revisiting Spatial Numerical Understanding in VLMs
Paper • 2605.23898 • Published • 7
Reward-Decomposed Reinforcement Learning for Immersive Video Role-Playing
Paper • 2605.04733 • Published
On-Policy Distillation with Best-of-N Teacher Rollout Selection
Paper • 2605.09725 • Published
Anisotropic Modality Align
Paper • 2605.07825 • Published • 27
CocoaBench: Evaluating Unified Digital Agents in the Wild
Paper • 2604.11201 • Published • 37
ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model
Paper • 2603.22281 • Published • 22
EgoForge: Goal-Directed Egocentric World Simulator
Paper • 2603.20169 • Published • 10
Vision Language Models Cannot Reason About Physical Transformation
Paper • 2603.07109 • Published • 2