Abstract:
The Multi Armed Bandit (MAB) problem in the Reinforcement Learning field refers to the problem of allocating resources in certain choices for achieving an overall goal with the objective of maximizing expected gain. Our BTP Journey is based on the goal of using MAB algorithms on hardware to effectively achieve better designs of hardware accelerators for various problems. In this semester we have focussed on learning about UCB (Upper confidence Bound) algorithm and its role in Early-Exiting of Deep Neural networks (with split computing) [SplitEE].
