Learning algorithms to preserve privacy and cluster bandits

This will be a two-part talk. In the first part, I will describe a method to sanitize datasets from containing information about sensitive attributes. This method uses adversarial networks to approximate a privacy-utility optimization problem. In the second part, I will describe a new objective in multi-armed stochastic bandits, that of clustering the arms based on their reward means. We will see the performance of UCB and Elimination style algorithms in attaining this objective.
November 13 @ 12:30
12:30 pm (1h)

Discovery Building

Ardhendu Tripathy