Summary of Regulating Model Reliance on Non-robust Features by Smoothing Input Marginal Density, By Peiyu Yang et al.
Regulating Model Reliance on Non-Robust Features by Smoothing Input Marginal Density
by Peiyu Yang, Naveed Akhtar, Mubarak Shah, Ajmal Mian
First submitted to arxiv on: 5 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 proposed framework regulates model reliance on non-robust features by attributing predictions to inputs. Robust feature attributions show consistency, while non-robust ones are susceptible to fluctuations. This allows identifying correlations between model reliance and smoothness of marginal densities. The gradients of the marginal density are regularized for robustness. An efficient implementation addresses numerical instability. In contrast to prevalent input gradient regularization, which smoothens conditional or joint densities, our method smoothes marginal densities. Experiments validate the effectiveness in addressing feature leakage and mitigating spurious correlations. The technique enables models to be robust against perturbations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper tries to make machine learning more trustworthy by making sure models don’t rely too heavily on features that can change easily. They create a way to see which features are important for the model’s predictions, and if those features are not very reliable, they try to fix it. This helps prevent models from being tricked into making wrong decisions because of faulty data. The researchers tested their method and found it worked well in stopping bad things from happening. |
Keywords
» Artificial intelligence » Machine learning » Regularization
