Combining Diverse Feature Priors

Abstract

To improve model generalization, model designers often restrict the features that their models use, either implicitly or explicitly. In this work, we explore the design space of leveraging such feature priors by viewing them as distinct perspectives on the data. Specifically, we find that models trained with diverse sets of feature priors have less overlapping failure modes, and can thus be combined more effectively. Moreover, we demonstrate that jointly training such models on additional (unlabeled) data allows them to correct each other's mistakes, which, in turn, leads to better generalization and resilience to spurious correlations. Code available at https://github.com/MadryLab/copriors

References in corpus (24)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition
• Shortcut Learning in Deep Neural Networks
• A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
• ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
• Certified Adversarial Robustness via Randomized Smoothing
• Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
• Invariant Risk Minimization
• Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning
• Towards Understanding Ensemble, Knowledge Distillation and Self-Distillation in Deep Learning
• Adversarial Examples Are a Natural Consequence of Test Error in Noise
• Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data
• In Defense of Pseudo-Labeling: An Uncertainty-Aware Pseudo-label Selection Framework for Semi-Supervised Learning
• Adversarial Robustness as a Prior for Learned Representations
• Noise or Signal: The Role of Image Backgrounds in Object Recognition
• Frustratingly Simple Domain Generalization via Image Stylization
• Informative Dropout for Robust Representation Learning: A Shape-bias Perspective
• Are Perceptually-Aligned Gradients a General Property of Robust Classifiers?
• Hyperparameter Ensembles for Robustness and Uncertainty Quantification
• Towards Shape Biased Unsupervised Representation Learning for Domain Generalization
• On the effectiveness of adversarial training against common corruptions
• Diverse Ensembles Improve Calibration
• Self-Adaptive Training: Bridging Supervised and Self-Supervised Learning
• Out-distribution aware Self-training in an Open World Setting