Toward Learning Human-aligned Cross-domain Robust Models by Countering Misaligned Features
arXiv:2111.03740
Abstract
Machine learning has demonstrated remarkable prediction accuracy over i.i.d data, but the accuracy often drops when tested with data from another distribution. In this paper, we aim to offer another view of this problem in a perspective assuming the reason behind this accuracy drop is the reliance of models on the features that are not aligned well with how a data annotator considers similar across these two datasets. We refer to these features as misaligned features. We extend the conventional generalization error bound to a new one for this setup with the knowledge of how the misaligned features are associated with the label. Our analysis offers a set of techniques for this problem, and these techniques are naturally linked to many previous methods in robust machine learning literature. We also compared the empirical strength of these methods demonstrated the performance when these previous techniques are combined, with an implementation available at https://github.com/OoDBag/WR
arXiv commentto appear at UAI 2022
References in corpus (18)
- mixup: Beyond Empirical Risk Minimization
- Improved Regularization of Convolutional Neural Networks with Cutout
- ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
- Domain Adaptation: Learning Bounds and Algorithms
- Beyond Sharing Weights for Deep Domain Adaptation
- Generalizing Across Domains via Cross-Gradient Training
- Bridging Theory and Algorithm for Domain Adaptation
- Measuring the tendency of CNNs to Learn Surface Statistical Regularities
- Natural Adversarial Examples
- Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs
- Does Distributionally Robust Supervised Learning Give Robust Classifiers?
- Biologically inspired protection of deep networks from adversarial attacks
- Learning Robust Representations by Projecting Superficial Statistics Out
- Learning from Failure: Training Debiased Classifier from Biased Classifier
- A New PAC-Bayesian Perspective on Domain Adaptation
- Don't Take the Easy Way Out: Ensemble Based Methods for Avoiding Known Dataset Biases
- Learning From Brains How to Regularize Machines
- The Struggles of Feature-Based Explanations: Shapley Values vs. Minimal Sufficient Subsets