LEARNING DYNAMICS - SYSTEM-IDENTIFICATION FOR PERCEPTUALLY CHALLENGEDAGENTS

From the perspective of an agent, the input/output behavior of the env ironment in which it is embedded can be described as a dynamical syste m. Inputs correspond to the actions executable by the agent in making transitions between states of the environment. Outputs correspond to t he perceptual information available to the agent in particular states of the environment. We view dynamical system identification as inferen ce of deterministic finite-state automata from sequences of input/outp ut pairs. The agent can influence the sequence of input/output pairs i t is presented by pursuing a strategy for exploring the environment. W e identify two sorts of perceptual errors: errors in perceiving the ou tput of a state and errors in perceiving the inputs actually carried o ut in making a transition from one state to another. We present effici ent, high-probability learning algorithms for a number of system ident ification problems involving such errors. We also present the results of empirical investigations applying these algorithms to learning spat ial representations.