Recent Advances in Meta and Self-Supervised Learning for Visual Analysis

Abstract

Deep learning has shown its success in various tasks in areas such as computer vision and natural language processing.
However, while deep learning models trained in fully-supervised fashions have exhibited promising performances, how to design and train such models with small or unsupervised data remains a challenging task. In this talk, I will go over recent trends for deep learning (particularly in computer vision) on the development of meta and self-supervised learning strategies, which can be applied to tackle the aforementioned challenging yet practical settings. I will also talk about our recent NeurIPS, CVPR, and ECCV works, and explain how we advance these learning schemes for visual analysis.


線上參與請點選連結

附件下載