Learning from Failure: Training Debiased Classifier from Biased Classifier

6 Jul 2020  ·  Junhyun Nam, Hyuntak Cha, Sungsoo Ahn, Jaeho Lee, Jinwoo Shin ·

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.

PDF Abstract

Datasets


Introduced in the Paper:

BAR

Used in the Paper:

CIFAR-10 MNIST CelebA ImageNet-W UrbanCars
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Action Recognition BAR LfF Accuracy 62.98 # 2
Out-of-Distribution Generalization ImageNet-W LfF (ResNet-50) IN-W Gap -17.57 # 1
Carton Gap +40 # 1
Out-of-Distribution Generalization UrbanCars LfF BG Gap -11.6 # 1
CoObj Gap -18.4 # 1
BG+CoObj Gap -63.2 # 1

Methods


No methods listed for this paper. Add relevant methods here