I'm working towards my Ph.D in computer science at Brandeis University. My advisor is Jordan Pollack, who runs the Demo Lab. The lab works on a wide variety of areas, including evolutionary computation, evolutionary robotics, and neural networks.

My research has come to focus on coevolutionary algorithm dynamics from the computer-science perspective of machine learning. From an optimization perspective, coevolutionary algorithms have been proposed as a principled way to allow the ability of the learner and the difficulty of the learning environment evolve together, without the need to inject human-supplied bias in the form of an engineered learning gradient or a pre-adapted learner. But, practice reveals this hypothesis to be somewhere in between fact and fiction---the many successes of coevolution are balanced by many irksome modes of "failure" that commonly recur. Investigation into these modes of failure has proven to be a challenge. While formal models of evolutionary algorithm dynamics have burgeoned, the relative lack of formal tools for the analysis of co-evolutionary algorithms has forced most investigations in the evolutionary computation community to be empirical in nature. The reason for this divide stems from the need to formally account for the defining characteristic of coevolution: the interaction of coevolving entities. I believe this property of coevolutionary systems is fundamentally game-theoretic in nature. My current research brings evolutionary game theory, mathematical biology, and theories of dynamical systems to bear on the question of coevolutionary algorithm dynamics.

More info: http://www.cs.brandeis.edu/~sevan