Summary of Disco: Discovering Overfittings As Causal Rules For Text Classification Models, by Zijian Zhang et al.
DISCO: DISCovering Overfittings as Causal Rules for Text Classification Models
by Zijian Zhang, Vinay Setty, Yumeng Wang, Avishek Anand
First submitted to arxiv on: 7 Nov 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL); 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 The paper introduces DISCO, a novel method for discovering global, rule-based explanations of neural language models’ decision-making processes. Existing post-hoc interpretability methods focus on unigram features of single input textual instances, failing to capture the models’ full decision-making process. DISCO employs a scalable sequence mining technique to extract relevant text spans from training data, associate them with model predictions, and conduct causality checks to distill robust rules that elucidate model behavior. These rules expose potential overfitting and provide insights into misleading feature combinations. The method is validated through extensive testing, demonstrating its superiority over existing methods in offering comprehensive insights into complex model behaviors. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new way to understand how neural language models work. Right now, these models are very good at understanding human language, but it’s hard for humans to figure out why they make certain decisions. The authors created a new method called DISCO that can explain the decision-making process of these models. DISCO looks at patterns in text and finds rules that explain how the model is making predictions. This helps identify potential problems with the model and provides insights into what it’s doing well or poorly. The paper shows that DISCO is better than existing methods for understanding complex language models. |
Keywords
» Artificial intelligence » Overfitting
