Research

In a nutshell, my main focus is on learning theory. I have studied some topics in online learning, mostly multi-armed bandit. I have also been working on adversarial perturbations.

I am also interested in theoretical computer science, mostly metric embedding; as well as mixed integer optimization.

Online Learning

We analyze the performance of online learning algorithm when the feedback structure is defined by a directed graph. The loss function is given by an adversary instead of a fixed distribution. We also give game-theoretic lower bound for online games when the adversary has different powers.

Temporal Reasoning

The basic task is to develop a method to extract the temporal clause like “this sunday” from the text, and normalize them to format resembles “2018-01-01”. The normalization format that most people in NLP community use is TIMEX3.

There are many higher-level tasks based on this block. For example, one could ask about the timeline of events in a newpaper, or causal relations between events. These tasks are generally hard due to the sparsity of the training data