Seminars

View all Seminars  |  Download ICal for this event

Differential Privacy in Multi-armed Bandits

Series: Bangalore Theory Seminars

Speaker: Sayak Ray Chowdhury Microsoft Research Bangalore

Date/Time: Aug 18 11:00:00

Location: CSA Seminar Hall (Room No. 254, First Floor)

Abstract:
Differential privacy (DP) has been recently introduced to multi-armed bandits to formally address the privacy concerns in its associated personalized services to participating users (e.g., recommendations). Prior work largely focusses on two trust models of DP �?? the central model, where a central server is responsible for protecting users sensitive data, and the (stronger) local model, where information needs to be protected directly on users side. However, there remains a fundamental gap in the utility achieved by learning algorithms under these two privacy models �?? the regret suffer in the local model is higher than that in the weaker central model.

In this talk, we will focus on a stronger model of trust than the central model, while suffering a smaller regret than the local model by considering recently popular shuffle and distributed model of privacy. We will demonstrate this privacy-regret trade-off for both K-armed bandit problem and linear contextual bandit problem.


Microsoft Teams link:

Link

We are grateful to the Kirani family for generously supporting the theory seminar series


Hosts: Rameesh Paul, Rachana Gusain, Rahul Madhavan, KVN Sreenivas