Speaker: Zhen-Yu Zhang, Nanjing University
Title: Towards Robust Semi-supervised Learning
Abstract: The semi-supervised learning paradigm offers an effective way to exploit large amounts of unlabeled data. Robustness is a central requirement of modern semi-supervised learning, since in many real-world applications the collected labeled data can be noisy and the learning environments can even change over time. This talk will introduce our recent efforts on this topic. Our study advances label noise tolerated semi-supervised learning and adaptive semi-supervised learning in changing environments. The proposed approaches enjoy nice theoretical guarantees and demonstrate real-world effectiveness.
