Learning De-biased Representations with Biased Representations

Abstract

Many machine learning algorithms are trained and evaluated by splitting data from a single source into training and test sets. While such focus on in-distribution learning scenarios has led to interesting advancement, it has not been able to tell if models are relying on dataset biases as shortcuts for successful prediction (e.g., using snow cues for recognising snowmobiles), resulting in biased models that fail to generalise when the bias shifts to a different class. The cross-bias generalisation problem has been addressed by de-biasing training data through augmentation or re-sampling, which are often prohibitive due to the data collection cost (e.g., collecting images of a snowmobile on a desert) and the difficulty of quantifying or expressing biases in the first place. In this work, we propose a novel framework to train a de-biased representation by encouraging it to be different from a set of representations that are biased by design. This tactic is feasible in many scenarios where it is much easier to define a set of biased representations than to define and quantify bias. We demonstrate the efficacy of our method across a variety of synthetic and real-world biases; our experiments show that the method discourages models from taking bias shortcuts, resulting in improved generalisation. Source code is available at https://github.com/clovaai/rebias.

arXiv commentAccepted to ICML 2020. Code available at https://github.com/clovaai/rebias

Cited by in corpus (26)

• AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights
• Learning from Failure: Training Debiased Classifier from Biased Classifier
• Learning Debiased Representation via Disentangled Feature Augmentation
• Can Subnetwork Structure be the Key to Out-of-Distribution Generalization?
• Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Models
• Which Shortcut Cues Will DNNs Choose? A Study from the Parameter-Space Perspective
• Evidential Deep Learning for Open Set Action Recognition
• Latent Adversarial Debiasing: Mitigating Collider Bias in Deep Neural Networks
• Benign Shortcut for Debiasing: Fair Visual Recognition via Intervention with Shortcut Features
• EnD: Entangling and Disentangling deep representations for bias correction
• What is Wrong with One-Class Anomaly Detection?
• Using Positive Matching Contrastive Loss with Facial Action Units to mitigate bias in Facial Expression Recognition
• Learning Debiased and Disentangled Representations for Semantic Segmentation
• Pulling Up by the Causal Bootstraps: Causal Data Augmentation for Pre-training Debiasing
• SWAD: Domain Generalization by Seeking Flat Minima
• BiaSwap: Removing dataset bias with bias-tailored swapping augmentation
• Interpreting Face Inference Models using Hierarchical Network Dissection
• StyleAugment: Learning Texture De-biased Representations by Style Augmentation without Pre-defined Textures
• Causal Attention for Unbiased Visual Recognition
• Unsupervised Learning of Debiased Representations with Pseudo-Attributes
• Invariant Representations with Stochastically Quantized Neural Networks
• RC2020 Report: Learning De-biased Representations with Biased Representations
• Combating Unknown Bias with Effective Bias-Conflicting Scoring and Gradient Alignment
• Fighting Fire with Fire: Contrastive Debiasing without Bias-free Data via Generative Bias-transformation
• Learning Robust Convolutional Neural Networks with Relevant Feature Focusing via Explanations
• Unsupervised Learning of Unbiased Visual Representations