Jacob Schrum's Page for Summer 2014

e-mail: schrum2@cs.utexas.edu

I've just successfully completed my Ph.D. in Computer Science at the University of Texas at Austin. This summer I will be moving to Georgetown, Texas to prepare for my new job as an Assistant Professor of Computer Science at Southwestern University, which is also where I received my undergraduate B.S. with a triple-major in Computer Science, Math, and German.

My dissertation advisor was Risto Miikkulainen of the Neural Networks Research Group. I'm interested in automatic discovery of complex multi-modal behavior, particularly in the domain of video games. Agents that can behave in different manners in response to different situations are crucial for games because they are so complex, and human players adapt so quickly. I'm particularly interested in the use of multiobjective evolution and neuroevolution in these domains. Furthermore, I am interested in finding domain-independent methods to solve these tasks, using tools such as fitness shaping. The less expert knowledge, the better.

The first half of my dissertation research used the BREVE simulation environment. My source code is available here. The ideas and code used in BREVE were extended in the second half of my dissertation, which focuses on the domain of Ms. Pac-Man using the Java implementation available here. I've developed a software framework in Java for evolving complex, multimodal behavior in this domain and others: Modular Multiobjective NEAT, or MM-NEAT, is an extension of the original NEAT algorithm that adds support for multiobjective evolution via NSGA-II, and modular neural networks that have separate modules for separate output policies. The fitness-shaping approach of Targeting Unachieved Goals (TUG), first introduced in this paper, is also supported.

I've also done some work in the domain of Unreal Tournament 2004 using the programming API Pogamut, which communicates with the game according to the Gamebots message protocol. Along with Igor Karpov, I participated in the annual Botprize and Humanlike Bots Competitions from 2008 to 2012, when our team finally won the grand prize by attaining a humanness rating of 51.9%. We also won the year's preceding Humanlike Bots Competition. Information about our work in Unreal Tournament is available on our Humanlike Bots Project Page. There are also some movies from the 2012 competition here. The source code for our bot is available here.

Over the summer, I'll mostly be preparing for the courses I will teach at Southwestern University in the Fall, but I will also be traveling to the Genetic and Evolutionary Computation Conference in July to present a paper that focuses on a portion of my dissertation research.


Journal Articles

Conference Publications

Book Chapters

Technical Reports


Previously taught classes (at UT): Previous TA work (at UT):


Last Updated: 5/13/14