Summary of Not Only the Last-layer Features For Spurious Correlations: All Layer Deep Feature Reweighting, by Humza Wajid Hameed et al.
Not Only the Last-Layer Features for Spurious Correlations: All Layer Deep Feature Reweighting
by Humza Wajid Hameed, Geraldin Nanfack, Eugene Belilovsky
First submitted to arxiv on: 23 Sep 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 This paper addresses the issue of spurious correlations in machine learning models, particularly when aiming for group-level fairness. By re-training the last layer on a balanced validation dataset, researchers have shown that robust features can be isolated for predictors. However, key attributes can sometimes be discarded by neural networks towards the last layer. To combat this, the authors consider retraining a classifier on a set of features derived from all layers. They utilize a feature selection strategy to select unbiased features from all the layers, observing significant improvements in worst-group accuracy on several standard benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps machine learning models be fairer by fixing mistakes they make when trying to predict certain groups. When training a model, sometimes it “forgets” important information that is important for predicting other groups. To fix this, the researchers tried retraining the model using all the information from previous layers. They used a special method to select only the most important features and found that this improved how well the model performed on certain groups. |
Keywords
» Artificial intelligence » Feature selection » Machine learning
