Rethinking Importance Weighting for Deep Learning under Distribution Shift

Abstract

Under distribution shift (DS) where the training data distribution differs from the test one, a powerful technique is importance weighting (IW) which handles DS in two separate steps: weight estimation (WE) estimates the test-over-training density ratio and weighted classification (WC) trains the classifier from weighted training data. However, IW cannot work well on complex data, since WE is incompatible with deep learning. In this paper, we rethink IW and theoretically show it suffers from a circular dependency: we need not only WE for WC, but also WC for WE where a trained deep classifier is used as the feature extractor (FE). To cut off the dependency, we try to pretrain FE from unweighted training data, which leads to biased FE. To overcome the bias, we propose an end-to-end solution dynamic IW that iterates between WE and WC and combines them in a seamless manner, and hence our WE can also enjoy deep networks and stochastic optimizers indirectly. Experiments with two representative types of DS on three popular datasets show that our dynamic IW compares favorably with state-of-the-art methods.

arXiv commentNeurIPS 2020 camera-ready version (this paper was selected for spotlight presentation)

References in corpus (14)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• A systematic study of the class imbalance problem in convolutional neural networks
• Domain Adaptation for Statistical Classifiers
• Learning to Reweight Examples for Robust Deep Learning
• Classification with Noisy Labels by Importance Reweighting
• Asymmetric Tri-training for Unsupervised Domain Adaptation
• Are Anchor Points Really Indispensable in Label-Noise Learning?
• Masking: A New Perspective of Noisy Supervision
• How does Disagreement Help Generalization against Label Corruption?
• Part-dependent Label Noise: Towards Instance-dependent Label Noise
• SIGUA: Forgetting May Make Learning with Noisy Labels More Robust
• On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data
• Mitigating Overfitting in Supervised Classification from Two Unlabeled Datasets: A Consistent Risk Correction Approach

Cited by in corpus (3)

• Learning with Instance-Dependent Label Noise: A Sample Sieve Approach
• DGMamba: Domain Generalization via Generalized State Space Model
• Enhanced Imitation Learning of Model Predictive Control through Importance Weighting