Summary of Hu at Semeval-2024 Task 8a: Can Contrastive Learning Learn Embeddings to Detect Machine-generated Text?, by Shubhashis Roy Dipta and Sadat Shahriar
HU at SemEval-2024 Task 8A: Can Contrastive Learning Learn Embeddings to Detect Machine-Generated Text?
by Shubhashis Roy Dipta, Sadat Shahriar
First submitted to arxiv on: 19 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper presents a system for detecting machine-generated text, specifically designed for the SemEval-2024 Task 8. The authors aim to address the limitations of previous detection systems that rely on knowing the specific text-generating model used. They propose a single contrastive learning-based model that uses approximately 40% fewer parameters than the baseline (149M vs. 355M) yet achieves comparable performance, ranking 21st out of 137 participants on the test dataset. The key finding is that a single base model can achieve similar performance using data augmentation and contrastive learning, without requiring an ensemble of multiple models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us detect when text has been generated by a machine, which is important because machines are creating fake texts to trick people. Lots of systems have tried to solve this problem, but they all rely on knowing the specific machine that made the text. This doesn’t work in real life because we often can’t figure out which machine was used. The authors came up with a new way to do it using something called contrastive learning. Their system uses fewer parameters than others and still works well. They found that one good model can be just as good as many models working together. |
Keywords
* Artificial intelligence * Data augmentation
