---
tags:
- text-classification
widget:
  - text: "This paper presents OASIS, novel one-pass aligned atlas set for image segmentation. Traditional atlas-based segmentation methods often require multiple iterations of registration, which can be time-consuming and computationally expensive. OASIS addresses this limitation by introducing one-pass alignment process that efficiently registers template atlas to a target image. This process involves two steps: coarse alignment using a deep convolutional neural network, followed by a fine alignment using a robust multiresolution registration algorithm. Experimental results on various medical imaging datasets demonstrate that OASIS achieves competitive segmentation accuracy compared to state-of-the-art methods, while significantly reducing computation time. Additionally, OASIS exhibits robustness against image artifacts and variations, making it suitable for wide range of applications in medical imaging and beyond. Overall, this paper presents a new approach to atlas-based image segmentation that addresses the limitations of traditional methods, and offers improved speed, accuracy, and robustness."
    example_title: example1
  - text: "High school graduation is often looked upon as a milestone for students to celebrate as the end of a long educational journey. However, some school districts have begun allowing students to graduate a year early. Though this can be beneficial in certain cases, there are several compelling reasons why this practice should not be made universally available to all high school students. Most importantly, graduates who jump the gun and finish high school early could be missing out on valuable learning experiences. High school is a formative period and the curriculum is designed to prepare young adults for college or the workforce. Skipping ahead by a year could lead to students not having enough guidance or mentorship available while learning vital skills and life lessons. In addition, early graduation can leave students feeling unprepared for the world outside of high school. Not having the same college exposure as other peers or lacking the maturity that comes with an extra year of school can hold graduates back from success in the long-run. Graduating a year earlier than expected does not always mean students will attend college earlier; in fact, it could mean starting a career ill-equipped to handle all the responsibilities that come with being a full-time working adult. For these reasons, schools districts should not make it possible for all high school students to graduate a year early. The benefits may be enticing, but there are many potential downsides to consider."
    example_title: example2
  - text: "Summer is a time of hanging out with friends at the beach and relaxing in the warm weather, but there is always something that swoops in to mess it up: a summer project. Summer projects are not all bad, for they are designed to ensure that students continue to learn during their three month getaway from school. However, who should design the summer project the student or the teacher? Summer projects should be teacher-designed because teachers have more experience with projects than students, and students will lean toward designing a quick and easy project. Teachers have had more encounters with school assigned projects in their lifetime than their students. Most teachers have college degrees. They have done the twelve plus years of primary school to graduate, and then gone on to college and done two to four more years to get their degree. Through all that schooling many projects have been assigned and completed. All of this gives teachers the proper experience and background to come up with a well designed summer project for their students. Just this past summer my computer system networking teacher, Mr. Generic_Name, assigned us a project. It was a review project incorporating all of the course material from my sophomore year that he did not want us to forget during the summer. I remember how well rounded of a project it was, so much so that I had to ask him how he come up with it. Mr. Generic_Name told me that the project was inspired by all the projects he encountered during his four years at William and Mary. Due to Mr. Generic_Name having plenty of experience with higher education projects, it allowed him to design a well developed summer project for my peers and I. I can only imagine the toddler like projects that would have been produced if my peers and I were allowed to design our summer projects instead of Mr. Generic_Name. More than likely if a student is given the opportunity to do less work they are going to take that opportunity. Throughout my years of high school I have seen this exact manner take place over and over. Students will take the easy way out by selecting courses that have a history of demanding very little from those enrolled. This same mannerism repeats itself once the student is enrolled in a course and it comes time to actually work. Just this past marking period my Math Analysis teacher, Mrs. Generic_Name, gave us a take home test for winter break. This test consisted of 47 problems. The more problems you did the higher grade you got, but students did not seem to take advantage of this. More than half of my class chose to only do 30 of the 47 problems. This was because completing 30 of the 47 problems earned the minimum for a passing grade, a 70 percent. My peers had full control of their grades and chose to only work for the minimum passing score so that the rest of their winter break would be free of math. This is what will occur if a student is allowed to design their own summer project. Once a student realizes that they get to design their project they will make it something simple, lousy, and childlike. The student will design something just good enough too pass, so that they can return to their more practical summer. Some argue that a student would be more inclined to do their project if they got to design it. However, this is not true as someone's interest in something does not determine whether they will do it or not, but their work ethic is the true determining factor. Summer projects serve a meaningful purpose; to make sure kids maintain learning and strengthening their brains even when school is out. Something with a strong purpose like that should not be taken lightly and should be handled by someone who can produce the best project. This is why summer projects should be teacher-designed, so that students can take away as much information as possible when the project is all said and done."
    example_title: example3
metrics:
- accuracy
- f1
- roc_auc
base_model:
- intfloat/e5-small
library_name: transformers
datasets:
- liamdugan/raid
model-index:
- name: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors
  results:
  - task:
      type: text-classification
    dataset:
      name: RAID-test
      type: RAID-test
    metrics:
    - name: accuracy
      type: accuracy
      value: 0.939
    source:
      name: RAID Benchmark Leaderboard
      url: https://raid-bench.xyz/leaderboard
pipeline_tag: text-classification
---

# My LoRA Fine-Tuned AI-generated Detector

This is a e5-small model fine-tuned with LoRA for sequence classification tasks. It is optimized to classify text into AI-generated or human-written with high accuracy.

- **Label_0**: Represents **human-written** content.
- **Label_1**: Represents **AI-generated** content.

## Model Details

- **Base Model**: `intfloat/e5-small`
- **Fine-Tuning Technique**: LoRA (Low-Rank Adaptation)
- **Task**: Sequence Classification
- **Use Cases**: Text classification for AI-generated detection.
- **Hyperparameters**: 
   - Learning rate: `5e-5`
   - Epochs: `3`
   - LoRA rank: `8`
   - LoRA alpha: `16`


## Training Details

- **Dataset**:
    - 10,000 twitters and 10,000 rewritten twitters with GPT-4o-mini.
    - 80,000 human-written text from [RAID-train](https://github.com/liamdugan/raid).
    - 128,000 AI-generated text from [RAID-train](https://github.com/liamdugan/raid).
- **Hardware**: Fine-tuned on a single NVIDIA A100 GPU.
- **Training Time**: Approximately 2 hours.
- **Evaluation Metrics**:
| Metric | (Raw) E5-small | Fine-tuned |
|--------|---------------:|-----------:|
|Accuracy| 65.2%          | 89.0%      |
|F1 Score| 0.653          | 0.887      |
| AUC    | 0.697          | 0.976      |

## Collaborators

- **Menglin Zhou**
- **Jiaping Liu**
- **Xiaotian Zhan**


## Citation
If you use this model, please cite the RAID dataset as follows:
```
@inproceedings{dugan-etal-2024-raid,
    title = "{RAID}: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors",
    author = "Dugan, Liam  and
      Hwang, Alyssa  and
      Trhl{\'\i}k, Filip  and
      Zhu, Andrew  and
      Ludan, Josh Magnus  and
      Xu, Hainiu  and
      Ippolito, Daphne  and
      Callison-Burch, Chris",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.674",
    pages = "12463--12492",
}
```