题目： Feature-based Transfer Learning via Kernel Embedding of Distributions
报告人： Sinno Jialin Pan 教授 南洋理工大学
摘要： Transfer learning is a learning paradigm motivated by human’s learning ability on transferring experience across different tasks or problems. Different from traditional machine learning paradigms, which assume training data and test data follow the same distribution, in transfer learning, training data and test data are usually drawn from different distributions or even represented in different feature spaces as they come from different domains or tasks. In this talk, I will first give an overview on transfer learning, and then present my work on feature-based transfer learning approaches through kernel embedding of distributions.