When does Bias Transfer in Transfer Learning?

Hadi Salman, Saachi Jain, Andrew Ilyas, Logan Engstrom, Eric Wong +1 more
7/6/2022
cs.LG

Official code: MadryLab/bias-transfer (15 stars, MIT, PyTorch, last commit 2022-07-24).

Abstract

Using transfer learning to adapt a pre-trained "source model" to a downstream "target task" can dramatically increase performance with seemingly no downside. In this work, we demonstrate that there can exist a downside after all: bias transfer, or the tendency for biases of the source model to persist even after adapting the model to the target class. Through a combination of synthetic and natural experiments, we show that bias transfer both (a) arises in realistic settings (such as when pre-training on ImageNet or other standard datasets) and (b) can occur even when the target dataset is explicitly de-biased. As transfer-learned models are increasingly deployed in the real world, our work highlights the importance of understanding the limitations of pre-trained source models. Code is available at https://github.com/MadryLab/bias-transfer

View on arXivView PDF

Code Implementations(1)

MadryLab/bias-transferOfficial100%
PyTorchMIT
150PythonJul 5, 20224 years agoMIT
biasdeep-learningpytorchtransfer-learning

Similar Papers

Overwriting Pretrained Bias with Finetuning Data

87% match1 repo52023

The Role of Pre-training Data in Transfer Learning

86% match1 repo102023

A Data-Based Perspective on Transfer Learning

86% match1 repo352022

Understanding Optimal Feature Transfer via a Fine-Grained Bias-Variance Analysis

85% match2024

Towards Foundation Models for Scientific Machine Learning: Characterizing Scaling and Transfer Behavior

85% match2026

Variation of Gender Biases in Visual Recognition Models Before and After Finetuning

84% match2023

Cite this paper

@article{salman2022when,
  title  = {When does Bias Transfer in Transfer Learning?},
  author = {Hadi Salman and Saachi Jain and Andrew Ilyas and Logan Engstrom and Eric Wong and Aleksander Madry},
  year   = {2022},
  eprint = {2207.02842},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2207.02842v1}
}

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/2207.02842.svg)](https://codeofpaper.com/paper/2207.02842)

Discussion