Fwd: PhD Final Oral Defense: Matthew Hausknecht, November 28, 2016 (Monday), 1:15pm, GDC 6.302

[Stephen Roller]

In case you all aren't aware

---------- Forwarded message ---------
From: Katie Dahm <kd...@cs.utexas.edu>
Date: Mon, Nov 14, 2016 at 10:09 AM
Subject: PhD Final Oral Defense: Matthew Hausknecht, November 28, 2016 (Monday), 1:15pm, GDC 6.302
To: CS-Grad-Students:; UTCS Graduate Students <gr...@cs.utexas.edu>
Cc: Gradoffice <grado...@cs.utexas.edu>

PhD Final Oral Defense: Matthew Hausknecht

Title: Cooperation and Communication in Multiagent Deep Reinforcement Learning

Date: November 28, 2016 (Monday)
Time: 1:15pm (CDT)
Place: GDC 6.302

Committee: Peter Stone (advisor), Ray Mooney, Dana Ballard, Risto Miikkulainen, Satinder Singh

Summary of thesis:
This thesis demonstrates that reinforcement learning combined with deep neural network function approximation can produce algorithms capable of discovering effective policies for domains with partial observability, parameterized-continuous actions spaces, and sparse rewards.

Additionally, we investigate architectures and algorithms suited for cooperative multiagent learning. We demonstrate that sharing parameters and memories between deep reinforcement learning agents fosters policy similarity, which can result in cooperative behavior. Additionally, we hypothesize that communication can further aid cooperation, and we present the Grounded Semantic Network (GSN), which learns a communication protocol grounded in the observation space and reward function of the task. In general, we find that the GSN is effective at helping agents learn to communicate on domains featuring partial observability and asymmetric information.