[Ml-stat-talks] Akshay Krishnamurthy talk on June 1st

Barbara Engelhardt bee at princeton.edu
Sat May 28 21:15:15 EDT 2016


June 1st 1:30pm-4:30pm, in Sherrerd 224.


On Sat, May 28, 2016 at 8:23 PM, Barbara Engelhardt <bee at princeton.edu>
wrote:

> Talk of interest, hosted by Han Liu:
>
> Title: Contextual-MDPs: A new model and analysis for reinforcement
> learning with rich observations
>
> Abstract:
>
> State-of-the-art reinforcement learning systems cannot solve hard problems
> because they lack systematic exploration mechanisms, instead relying on
> variants of uniform exploration to collect diverse experience. Systematic
> exploration mechanism exist for Markov Decision Processes (MDPs) but have
> sample complexity that scales polynomially with the number of unique
> observations, making them intractable for modern reinforcement learning
> applications where observations come from a visual sensor. Are there
> reinforcement learning algorithms that can effectively handle rich
> (high-dimensional, infinite) observations by engaging in systematic
> exploration? To aid in the development of such an algorithm, I will first
> describe a new model for reinforcement learning with rich observations that
> we call the Contextual-MDP. This model generalizes both stochastic
> contextual bandits and MDPs, but is considerably more tractable than
> Partially Observable Markov Decision Processes (POMDPs). I will then
> describe a new algorithm for learning optimal behavior in Contextual-MDPs.
> This algorithm engages in global exploration while using a function class
> to approximate future performance and has a sample complexity that scales
> polynomially in all relevant parameters while being independent of the
> number of unique observations. This represents an exponential improvement
> on all existing approaches and provides theoretical justification for
> reinforcement learning with function approximation.
>
> Bio
>
> Akshay Krishnamurthy is a postdoctoral researcher at Microsoft Research,
> New York City. In June 2015, he completed his PhD in the Computer Science
> Department at Carnegie Mellon University, advised by Aarti Singh. His
> research focuses on interactive learning and learning settings involving
> feedback-driven data collection, including reinforcement learning,
> interactive solutions for discovering low-dimensional structure, and
> complex prediction problems with limited feedback. He has also worked on
> problems in nonparametric statistics, anomaly detection, and compressive
> sensing. Starting Fall 2016, he will join the Department of Computer
> Science at the University of Massachusetts, Amherst as an assistant
> professor.
>
>


-- 
Barbara E Engelhardt
Assistant Professor
Department of Computer Science
Center for Statistics and Machine Learning
Princeton University
http://www.cs.princeton.edu/~bee
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20160528/9130f5ba/attachment.html>


More information about the Ml-stat-talks mailing list