Learning from Failure: Training Debiased Classifier from Biased Classifier

TL;DR

This paper introduces a novel debiasing method that trains a pair of neural networks where one is intentionally biased and the other is trained to counteract this bias, improving robustness against dataset biases.

Contribution

The paper proposes a generic, failure-based debiasing scheme using a pair of networks, one amplified with bias and the other trained to oppose it, without needing explicit bias labels.

Findings

Significant performance improvements on synthetic and real-world datasets.

Outperforms some methods requiring explicit bias supervision.

Effective against various bias types.

Abstract

Neural networks often learn to make predictions that overly rely on spurious correlation existing in the dataset, which causes the model to be biased. While previous work tackles this issue by using explicit labeling on the spuriously correlated attributes or presuming a particular bias type, we instead utilize a cheaper, yet generic form of human knowledge, which can be widely applicable to various types of bias. We first observe that neural networks learn to rely on the spurious correlation only when it is "easier" to learn than the desired knowledge, and such reliance is most prominent during the early phase of training. Based on the observations, we propose a failure-based debiasing scheme by training a pair of neural networks simultaneously. Our main idea is twofold; (a) we intentionally train the first network to be biased by repeatedly amplifying its "prejudice", and (b) we debias the training of the second network by focusing on samples that go against the prejudice of the biased network in (a). Extensive experiments demonstrate that our method significantly improves the training of the network against various types of biases in both synthetic and real-world datasets. Surprisingly, our framework even occasionally outperforms the debiasing methods requiring explicit supervision of the spuriously correlated attributes.