My primary research focus is on online learning with evolutionary algorithms. Currently I’m working with XCS, a rule based evolutionary algorithm proposed by Stewart W. Wilson. XCS is a learning classifier system that works under a reinforcement learning paradigm, and uses genetic algorithms to evolve maximally accurate and general populations of rules that correctly represent a concrete problem. XCS has been succesfully used to solve multi-step problems (problems where an agent learns from its interaction with the world), data mining problems, and function approximation problems.

I am also working with UCS, a learning classifier system derived from XCS, but specified to be applied only to classification tasks. UCS inherits its main components from XCS, but removes the update of parameters by reinforcement for an online estimation of the accuracy—i.e., percentage of correct classifications.