Online Seminar Computer Science - Dr. P. Gajane, Eindhoven University

Title: Sequential Decision Making with (Practically Motivated) Non-standard Feedback, and Fairness and Privacy in Machine Learning

Abstract:

In this presentation, I will mainly discuss the following two centralthemes of my research and their application in practical domains.

• Sequential decision making with non-standard feedback.

• Fairness and privacy in machine learning.

In both these themes, problems with non-stationary environments are of key interest as many practical applications feature non-stationary environments.

Firstly, I will present some of my research on sequential decision making problems with non-standard feedback. I will briefly explain the practical motivations for these non-standard forms of feedback and outline the novel efficient algorithms that we devised for these problems. I will also present the best-in-class theoretical performance guarantees that we proved for these algorithms. Next, we will see my research on efficient learning in non-stationary environments. Then, I will describe my research on human issues like privacy and fairness in machine learning. I will present our novel efficient algorithms which guarantee a strict notion of privacy. I will also illustrate the best-in-class theoretical performance guarantees that we proved for these algorithms.

Next, I will briefly explain how biased/unfair behavior by machine learning algorithms has been documented in high-impact domains. Further, I will present our work which shows a pathway for an interdisciplinary approach to fair machine learning. In the end, I will mention some of my interdisciplinary research on other diverse topics.