Shreya Sinha Roy
I am a 3rd year PhD student working under the supervision of Dr. Ritabrata Dutta, Dr. Richard Everitt, and Prof Christian Robert. My primary research interests are Computational Statistics and Reinforcement Learning. Currently, I am working on Bayesian Reinforcement Learning - exploring several Sampling Techniques for Model-based Reinforcement Learning.
Bayesian Deep Generative Reinforcement Learning:
Bayesian deep RL: The above diagram shows an episodic posterior update followed by a policy update routine for th episode
. We have assumed the episode length to be
, and
denotes the prior on the model parameters. Prequential Scoring Rule (
) is used to compute the generalized posterior which is based on true interaction data(
) and a simulation(
) from the deep generative model
. We obtain the samples,
from the generalized posterior,
via Sequential Monte Carlo (SMC) samplers. These samples are used to simulate
trajectories of interaction
from the model
. Optimal policy is trained by maximizing the averaged value of
function computed from the
simulated trajectories. The new policy,
is then used to interact with the true Environment in the next episode.
Publication:
Sinha Roy, S., Everitt, R., Robert, C., Dutta, R., "Bayesian Deep Generative Reinforcement Learning", to be submitted to NeurIPS, 2024.
Contact
Shreya.Sinha-Roy@warwick.ac.uk