I am interested in both artificial intelligence and systems. On the artificial intelligence side, I am particularly interested in evolutionary methods for developing intelligent behavior. On the systems side, I am most interested in distributed systems, and especially non-proprietary, dynamically connected systems. I am also interested in human-computer interaction and community-system interaction in terms of how these interactions can accelerate research efforts in AI and systems.
My interest in multi-agent systems ties the areas of AI and systems together. I am particularly interested in the complex behavior that arises when systems of intelligent agents interact. Distributed multi-agent systems are interesting in their own right, but I am intrigued far more by the research challenge of developing multi-agent behavior using distributed computation.
At Cornell University, I have completed four years of undergraduate study in the computer science honors program, and I am currently enrolled in a one-year engineering Master's program. I have finished two semester-long independent research projects in graphics and artificial intelligence. By the end of this academic year, I will have completed two more semester-long independent research projects and six Ph.D.-level courses, including study in theory, algorithms, systems, vision, and artificial intelligence. I have also built a multi-disciplinary background, with work in philosophy, art, music, English, and psychology, as well as the sciences.
My current research at Cornell is focused on using genetic algorithms for evolving neural networks to play the board game Go. I have investigated several network architectures and crossbreeding procedures, which has led me to develop a novel crossbreeding procedure for fixed-architecture networks. I have also been actively working on an extra-curricular distributed systems research project since May of this year. I have developed an indexed, distributed file system that features fast, exhaustive searches and modest networking requirements. As part of this project, I have investigated how node specialization can boost overall system performance above that which is achievable using a general peer-to-peer node model.
Continuing with research in these areas as a Ph.D. student, I would like to explore multi-agent interactions in the context of distributed systems. Specifically, I want to investigate spontaneously connected systems and adaptive agent protocols within such systems. With nondeterministic protocols that make probabilistic guarantees of correctness, I think that adaptive protocols (those that can be altered or that alter themselves substantially during operation) can be practical, especially in cases where the optimal protocol involves solving a provably hard problem. I plan to explore how a system can evaluate adaptive protocol performance at runtime for the purpose of building a population of agents that work well together. Among other possible avenues, I will look into how voting can be used to allow agents to evaluate each other's protocols.
Part of the focus of my graduate study will be on honing my teaching skills. I was an undergraduate teaching assistant for CS481, Cornell's honors automaton theory course. Currently, I am completing two semesters of additional work as a TA. I have found teaching to be a rewarding experience and also useful for solidifying my understanding of the material taught.
As a TA this fall for CS472, Artificial Intelligence, I developed a code framework for a class programming assignment. The framework supports multi-agent capture-the-flag games on two-dimensional maps of obstacles and allows students to write code to control the behavior of each agent on their team. The assignment asked students to work with multi-agent behavior in a limited-information environment, but it was open-ended so that any approach for behavior control and navigation was possible (many teams stuck with a heuristic search approach, while some students went as far as developing a complete neural network training system to generate good agent parameters). The framework also supports class-wide capture-the-flag tournaments and automatic generation of agent score and statistic reports. In the next few weeks, I will finalize the framework and make it available for use by AI courses at other schools. This programming assignment managed to motivate many students and to rejuvenate their interest in course material towards the end of the semester. As a graduate student, I plan to continue work on the problem of maintaining class motivation.
My long term goals involve teaching at a university that is focused on both education and research. I feel that I will be able to make a major contribution to research work in the areas of AI and systems, and I also hope to make a more general contribution to the academic world as a good teacher.