Please note: This seminar will be given online.
Weihao Kong, Postdoctoral researcher
Department of Computer Science, University of Washington
In this talk, I will discuss several examples of my research that reveal a surprising ability to extract accurate information from modest amounts of data.
First, I'll discuss the problem of learning from a large number of heterogeneous data sources: A common modern machine learning scenario involves a large amount of data contributed by a large number of heterogeneous individuals, with each individual providing a modest amount of data. In such a scenario, it is crucial the model being trained for each individual leverages data from other individuals (who may have different underlying models). The fundamental question is: To what extent can a large number of data sources compensate for the lack of data from each source? We answer this question in two fundamental settings where each data source is 1) a binomial distribution, 2) a linear model, by providing optimal algorithms and matching lower bounds.
In the second part of the talk, I'll discuss estimating learnability: Without enough data to learn a good model for prediction, is it possible to tell whether a good model exists? Surprisingly, we showed this is possible under linear model assumptions in supervised learning and contextual bandits settings. I will also discuss the implications of our techniques to model selection in this contextual bandits/reinforcement-learning setting.
Bio: Weihao Kong is currently a postdoc researcher at the University of Washington advised by Sham Kakade. He received his Ph.D. from the Computer Science Department at Stanford University in 2019 advised by Gregory Valiant. He received his B.S. from Shanghai Jiao Tong University in 2013. His research interests span machine learning, high-dimensional statistics, and theoretical computer science.
To join this seminar on Zoom, please go to https://zoom.us/j/92673241876?pwd=RkZSQzY5czRPM3ZXemhYVVZobjFPdz09.
200 University Avenue West
Waterloo, ON N2L 3G1