Summary of Is the Mmi Criterion Necessary For Interpretability? Degenerating Non-causal Features to Plain Noise For Self-rationalization, by Wei Liu et al.
Is the MMI Criterion Necessary for Interpretability? Degenerating Non-causal Features to Plain Noise for Self-Rationalization
by Wei Liu, Zhiying Deng, Zhongyu Niu, Jun Wang, Haozhao Wang, YuanKai Zhang, Ruixuan Li
First submitted to arxiv on: 8 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 proposes a novel criterion for extracting crucial rationales from input data, addressing the issue of spurious features in datasets that can hinder the discovery of causal relationships. The current state-of-the-art method, maximum mutual information (MMI), is prone to being misled by non-causal correlations between features and labels. To overcome this limitation, the paper introduces a new criterion called Maximizing Remaining Discrepancy (MRD), which treats spurious features as plain noise. The MRD approach allows models to learn from datasets with rich spurious features as if they were clean, improving rationale quality by up to 10.4% compared to competitive MMI variants. The paper demonstrates the effectiveness of MRD on six widely used datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper tries to solve a big problem in artificial intelligence. Right now, it’s hard for computers to understand why they made certain decisions. This is important because we want machines to learn from their mistakes and become better over time. One way to do this is by looking at the reasons behind a computer’s decision. However, there are many things that can make this process harder than it needs to be. For example, some data might not be related to what we’re trying to understand at all. The paper proposes a new way of solving this problem that works better than current methods. |
