Model of a plausible learning scheme for CA3 Hippocampus

Abstract

There is ample evidence that mammalian Hippocampus, defined as ammon's horn sectors 1 through 4, can learn. In particular, sector CA3 is presumed able to learn to produce a wide range of appropriate output patterns in response to extrinsic inputs arriving over the mossy fiber, temporo-ammonic, septo-hippocampal, and commissural systems. The neuroanatomy and neurophysiology of CA3 have been studied extensively, but the information conveyed by its signals and the way its circuit actions represent decisionary and memory functions remain elusive. Several roles have been posited for the Hippocampus in the guidance of animal behavior, but they are not inter-related in any clear way, and seem separated by several levels of abstraction from the circuit actions which actually embody decisionary and memory functions. We present a plastic circuit model of the CA3 sector and discuss its computer simulation as a caricature of developmental and neurophysiological processes. In particular, we concentrate on plausible ways in which excitatory and inhibitory interneurons can learn to properly influence articulate pyramidal cell output patterns under the guidance of only diffuse positive or negative reinforcement. Statistical learning in interneuronal-to-pyramidal synapses is posited to occur in one cell cluster after another until the overall circuit matures. A kind of intra-pyramidal Markovian learning is also posited and simulated, with quite remarkable results. A list of critical assumptions each of which is testable in principle is given, so the model can be thoroughly tested. The effect of each assumption is documented by computer simulation data.