Learning environmental clues in the KIV model of the cortico-hippocampal formation

11 Previous studies on the KIV model outlined a general architecture of modeling sensory– perceptual–intentional action cycle in the primordial vertebrate forebrain using nonlinear dy13 namical principles. KIV consists of three KIII units representing aperiodic/chaotic dynamics in sensory cortex, hippocampal formation, and midline forebrain, respectively. The sensory cortex 15 has demonstrated excellent performance as a pattern recognition and classi cation device. In this work, the behavior of the hippocampal formation is studied as part of the KIV system. 17 We elaborate a reinforcement algorithm to learn goal-oriented behavior based on global orientation beacons, biased by local sensory information provided by visual or infra-red sensors. We 19 illustrate the operation of the KIV model using the multiple T-maze navigation problem. c © 2004 Published by Elsevier B.V. 21