Dustin Richwine

Email: commonphilosophe@hotmail.com, alekhine@eden.rutgers.edu, dustinr@dimax.rutgers.edu

Office: CORE-336

Office Hours: By Appointment

School: Rutgers University, New Brunswick

Major: Mathematics

 


DIMACS REU Project

Faculty Mentor: Dr. Michael Littman, Professor and Chair of the Computer Science Department

Wikipedia

Homepage

Graduate Advisor: Michael Wunder

Homepage

Colleague: Christopher Kleven, Central College

DIMACS Page

Project Name: In Search of Value Equilibria

Project Description:

We intend to generalize on the results of previous game theoretic research on learning algorithms by analyzing self-play in environments of generalized numbers of players, states, and available actions. We hope to use information gathered through this analysis to design a new or improved learning algorithm for playing games. We are interested in determining the similarity of a successful learning algorithm's behavior to a natural learning algorithm's, or organism's behavior.

Slide Show

Relevant Papers:

Wunder's Q-learning Self-play Analysis

Singh's IGA Proof

Littman's Learning in Zero-Sum Games

Learning with Memory

Equilibria with Memory

Project Log

Week One (6/2-6/4):

Red "Classes of Multiagent Q-Learning Dynamics with Epsilon-Greedy Exploration". Red chapters 4 and 7 of Multiagent Systems: Algorithmic, Game Theoretic, and Logical Foundations. Red some of Theory of Games and Economic Behavior. Brainstormed on characteristics of an idealized learning algorithm.

Week Two (6/7-6/11):

Red "Nash Convergence of Gradient Dynamics in General-Sum Games" and "Markov Games as a Framework for Multi-Agent Reinforcement Learning". Red more of Theory of Games and Economic Behavior. Made website. Gave introductory presentation. Brainstormed on an idea for an idealized game-playing algorithm.

Week Three (6/14-6/18):

Coded a Q-Learner in as general a setting as possible using the common java language as well as its environment including a way to read a game in from a text file and play a Q-Learner against a Q, random-action, or human opponent (Code will soon be viewable below under "My Algorithms"). Red some of "AWESOME: A general multiagent algorithm that converges in self-play and learns a best response against stationary opponents" and brainstormed about the most desireable self-play characteristics of a game-playing algorithm designed to become openly popular.

Week Four (6/21-6/25):

Continued coding java environment. Researched ways to make a Graphic User Interface.

Week Five (6/28-7/2):

Learned GUI techniques with Javax.swing through the NetBeans IDE. Began coding GUI. Abandoned GUI idea. Considered methods of allowing humans to experimentally play games against algorithms.

My Algorithms

Q-Learner Game Player PlayImplementor

About Me

At Rutgers, I have experience in a variety of fields including math, economics, philosophy, and cognitive science. Within mathematics, I am especially interested in logic and foundations, theoretical computer science, game theory, probability and combinatorics, and all discrete math. My life goals are to learn as much as possible then to accumulate wealth with which to improve societal efficiency and quality of life through structural governmental changes. My philosophy is epistemically foundationalistic, metaphysically epiphenominalistically dualistic as well as atheistic, and ethically egoistic. My political views are fiscally conservative and socially liberal. I play chess and am learning baduk (go, weiqi). Comedy and Rhetoric are the only art forms I appreciate. I am a pessimist and apparently an introverted, intuitive, thinking, and judging person like most college professors. I have one younger brother. I live in Southern New Jersey, but am originally from northeastern Pennsylvania. I hope to attend graduate school in Economics in some state other than New Jersey.


Last modification 7/08/2010