Improving out-of-distribution generalization via multi-task self-supervised pretraining

• DOI: 10.48550/arxiv.2003.13525
• Publication Date: 2020-01-01
• Authors: Isabela Albuquerque; Nikhil Naik; Junnan Li; Nitish Shirish Ke...; Richard Socher
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; 03 medical and health sciences; 0302 clinical medicine; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; Machine Learning (cs.LG)

Self-supervised feature representations have been shown to be useful for supervised classification, few-shot learning, and adversarial robustness. We show that features obtained using self-supervised learning are comparable to, or better than, domain generalization in computer vision. introduce a new pretext task of predicting responses Gabor filter banks demonstrate multi-task compatible tasks improves performance as compared training individual alone. Features learnt through self-supervision obtain unseen domains when their counterpart there is larger shift between test distributions even localization ability objects interest. can also combined with other methods further boost performance.