Tuesday, April 03, 2018. 12:00PM. NSH 1507.

Back to Seminar Schedule

Han Zhao -- Multiple Source Domain Adaptation with Adversarial Learning

Abstract: While domain adaptation has been actively researched, most theoretical results and algorithms focus on the single-source-single-target adaptation setting. In the first part of the talk, I will discuss new generalization bounds for domain adaptation when there are multiple source domains with labeled instances and one target domain with unlabeled instances. The theory also leads to an efficient learning strategy using adversarial neural networks: I will show how to interpret it as learning feature representations that are invariant to the multiple domain shifts while still being discriminative for the learning task.

In the second part, I will discuss two models for multisource domain adaptations: the first model optimizes the worst-case bound, while the second model is a smoothed approximation of the first one and optimizes a task-adaptive bound. We also demonstrate the effectiveness of both models by conducting extensive experiments showing superior adaptation performance on three real-world datasets: sentiment analysis, digit classification, and vehicle counting.

This talk includes joint work with Shanghang Zhang, Guanhang Wu, Joao Costeira, Jose Moura and Geoff Gordon.