Mitigating Simplicity Bias in Deep Learning for Improved OOD Generalization and Robustness

Bhavya Vasudeva, Kameron Shahabi, Vatsal Sharan
10/9/2023
cs.LGstat.ML

Official code: estija/CMID (2 stars, MIT, last commit 2025-02-04).

Abstract

Neural networks (NNs) are known to exhibit simplicity bias where they tend to prefer learning 'simple' features over more 'complex' ones, even when the latter may be more informative. Simplicity bias can lead to the model making biased predictions which have poor out-of-distribution (OOD) generalization. To address this, we propose a framework that encourages the model to use a more diverse set of features to make predictions. We first train a simple model, and then regularize the conditional mutual information with respect to it to obtain the final model. We demonstrate the effectiveness of this framework in various problem settings and real-world applications, showing that it effectively addresses simplicity bias and leads to more features being used, enhances OOD generalization, and improves subgroup robustness and fairness. We complement these results with theoretical analyses of the effect of the regularization and its OOD generalization properties.

View on arXivView PDF

Code Implementations(1)

estija/CMIDOfficial
MIT

Code for the Conditional Mutual Information-Debiasing (CMID) method.

21PythonJun 16, 20231 years agoMIT
conditional-mutual-informationdebiasingdeep-learningdistribution-shiftood-generalization+1 more

Similar Papers

Simplicity Bias of Two-Layer Networks beyond Linearly Separable Data

85% match2024

Towards Robust Out-of-Distribution Generalization: Data Augmentation and Neural Architecture Search Approaches

84% match2024

A Closer Look at Model Adaptation using Feature Distortion and Simplicity Bias

84% match2023

Understanding and Improving Feature Learning for Out-of-Distribution Generalization

84% match1 repo292023

Principled Out-of-Distribution Generalization via Simplicity

84% match2025

Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization

83% match2024

Cite this paper

@article{vasudeva2023mitigating,
  title  = {Mitigating Simplicity Bias in Deep Learning for Improved OOD Generalization and Robustness},
  author = {Bhavya Vasudeva and Kameron Shahabi and Vatsal Sharan},
  year   = {2023},
  eprint = {2310.06161},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2310.06161v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2310.06161.svg)](https://codeofpaper.com/paper/2310.06161)

Discussion