Simplicity Bias

Simplicity bias, the tendency of machine learning models to favor simpler solutions over more complex, potentially more accurate ones, is a significant area of research. Current investigations focus on understanding this bias in various architectures, including neural networks (especially two-layer ReLU networks and transformers) and its impact on generalization, robustness, and fairness, often employing techniques like Sharpness-Aware Minimization or regularization strategies to mitigate its effects. Addressing simplicity bias is crucial for improving model performance, particularly in out-of-distribution settings and for preventing the amplification of existing biases in data, leading to more reliable and equitable AI systems.

Papers

May 25, 2023

Which Features are Learnt by Contrastive Learning? On the Role of Simplicity Bias in Class Collapse and Feature Suppression
Yihao Xue, Siddharth Joshi, Eric Gan, Pin-Yu Chen, Baharan Mirzasoleiman
Contrastive Learning Representation Learning Class Imbalance Feature Wise Simplicity Bias Class Representation Class Specific Feature Feature Suppression

March 23, 2023

A Closer Look at Model Adaptation using Feature Distortion and Simplicity Bias
Puja Trivedi, Danai Koutra, Jayaraman J. Thiagarajan
Native Robustness Distribution Generalization Glance Annotation Distortion Aware Model Adaptation Robust Generalization Simplicity Bias Effective Transfer Learning Adaptation Protocol

February 7, 2023

Delving Deep into Simplicity Bias for Long-Tailed Image Recognition
Xiu-Shen Wei, Xuhao Sun, Yang Shen, Anqi Xu, Peng Wang, Faen Zhang
Self Supervised Learning Generalization Performance Simplicity Bias Diving Deep Long Tailed Image Recognition

February 1, 2023

Simplicity Bias in 1-Hidden Layer Neural Networks
Depen Morwani, Jatin Batra, Prateek Jain, Praneeth Netrapalli
Neural Network Ensemble Approach Robust Classifier Simplicity Bias

January 30, 2023

Overcoming Simplicity Bias in Deep Networks using a Feature Sieve
Rishabh Tiwari, Pradeep Shenoy
Deep Network Spurious Feature Simplicity Bias

December 13, 2022

Simplicity Bias Leads to Amplified Performance Disparities
Samuel J. Bell, Levent Sagun
Model Bias Balanced Dataset Simplicity Bias Significant Disparity Performance Disparity

November 22, 2022

Simplicity Bias in Transformers and their Ability to Learn Sparse Boolean Functions
Satwik Bhattamishra, Arkil Patel, Varun Kanade, Phil Blunsom
Transformer Megatron Decepticons Generalization Performance Recurrent Network Critique Ability Simplicity Bias Random Transformation Sparse Function

November 21, 2022

Neural networks trained with SGD learn distributions of increasing complexity
Maria Refinetti, Alessandro Ingrosso, Sebastian Goldt
Neural Network Deep Neural Network Stochastic Gradient Descent Complexity Matter Simplicity Bias Order Statistic

October 4, 2022

Learning an Invertible Output Mapping Can Mitigate Simplicity Bias in Neural Networks
Sravanti Addepalli, Anshul Nasery, R. Venkatesh Babu, Praneeth Netrapalli, Prateek Jain
Neural Network LeArning Abstract Deep Neural Network Simplicity Bias Dynamic Feature Feature Reconstruction Invertible Linear Transformation Linear Learner

February 11, 2022

Improving Generalization via Uncertainty Driven Perturbations
Matteo Pagliardini, Gilberto Manunza, Martin Jaggi, Michael I. Jordan, Tatjana Chavdarova
Strong Generalization Stochastic Gradient Descent Simplicity Bias Uncertain System Weight Perturbation

December 22, 2021

A Simplicity Bubble Problem in Formal-Theoretic Learning Systems
Felipe S. Abrahão, Hector Zenil, Fabio Porto, Michael Winter, Klaus Wehmuth, Itala M. L. D'Ottaviano
Machine Learning Information Theory Statistical Learning Algorithmic Bias Dataset Generation Simplicity Bias