Lin Gong
Department of Computer Science
University of Virginia

• Date: Wednesday March 27th, 2019
• Time: 1:00PM
• Location: Rice 242

Abstract The advent of participatory web has created massive amounts of user-generated data, which enables the study of online user attributes and behaviors. Traditional social psychology studies commonly conduct surveys and experiments to collect user data in order to infer attributes of individuals, which are expensive and time-consuming. In contrast, we aim to understand users by building computational user models automatically, thereby to save time and efforts. And the principles of social psychology serve as good references for building such computational models.

In this presentation, I will discuss about my PhD works for modeling online user behaviors based on user-generated data. In particular, two challenges are addressed: (1) modeling users’ diverse ways of expressing attitudes or opinions; (2) building unified user models by integration of different modalities of user-generated data. This presentation bridges the gap between social psychology and computation behavior modeling which also provides a foundation for making user behavior modeling useful for many other applications.

Contextual Bandits in a Non-Stationary Environment

Huazheng Wang
Department of Computer Science
University of Virginia

• Date: Wednesday March 20th, 2019
• Time: 1:00PM
• Location: Rice 242

Abstract Multi-armed bandit algorithms have become a reference solution for handling the explore/exploit dilemma in recommender systems, and many other important real-world problems, such as display advertisement. However, such algorithms usually assume a stationary reward distribution, which hardly holds in practice as users’ preferences are dynamic. In this talk, I will introduce three works on non-stationary contextual bandit algorithms, including 1) detecting possible changes of environment, 2) finding a dynamic ensemble of admissible bandit models and 3) modeling user dependency for collaborative learning. Theoretical analysis and empirical evaluations on real-world recommendation datasets validate the effectiveness of the algorithms.

Building Fair Representations for Images and Text

Vicente Ordonez
Department of Computer Science
University of Virginia

• Date: Wednesday March 6th, 2019
• Time: 1:00PM
• Location: Rice 242

Abstract There has been an increased amount of attention on the societal consequences of deploying machine learning models for decision making. Machine learning models often make disparate decisions for different segments of the input population that correlate with protected variables such as race, gender, and age. A more problematic scenario in current systems, is their reliance on generic representations learned from large amounts of data and re-used for many downstream applications, e.g. pre-trained CNNs for images, and pre-trained word-embeddings for text. In our group, we have been studying problems of representation, where the objective is to build generic representations for images and text that do not encode information about protected variables. I will discuss some of our current work in this area in the domain of both images and text.

Deep Learning for Genomics

Jack Lanchantin
Department of Computer Science
University of Virginia

• Date: Wednesday February 27th, 2019
• Time: 1:00PM
• Location: Rice 242

Abstract Since the Human Genome project was completed in 2003, scientists have sought to discover the mechanisms by which DNA and its surrounding elements control biological processes. In this talk, I will introduce two essential biological processes related to the Genome and explain our work in using deep learning to predict and understand them. I will also discuss some of the open problems related to machine learning in Genomics.