HEBB MARR NETWORKS THE HIPPOCAMPUS AND SPATIAL MEMORY

点击翻译按钮获取中文摘要

Our goal is to develop a computational and empirical framework through which to understand the neuronal dynamics of the mammalian hippocampus and its cortical connections, in relation to spatial representation and memory. This system is crucial for memory formation, and its dysfunction through injury or disease is the major cause of human memory impairment. The project complements ongoing neurophysiological studies of single cell activity in freely behaving animals, spontaneous and evoked synaptic plasticity and modulation in conscious animals, and the micro-physiology of neural interaction using spike-triggered quantal analysis at single synapses in vitro. The wealth of data generated by such studies (from our own and other groups) has created an "embarrassment of riches" crisis. We plan to exacerbate the crisis by developing methods for recording from parallel arrays of "stereotrode" probes to acquire data from interacting populations of about 50 to 100 hippocampal neurons during spatial behavior. We hope to help alleviate the crisis thorough simulations that incorporate enough of the known network, biophysical and unit activity parameters that useful insights into the physiological origins of several important "system-level" phenomena can be derived and tested in physiological experiments. The conceptual starting point for these simulations is the Hebb-Marr formalism for distributed associative memory. Specifically, we address the origins and computational utility of the following phenomena: 1) Hippocampal neurons are exquisitely selective for the animal's location and orientation in space. Such selectivity is present neither at the inputs, nor (surprisingly) at the cortical outputs of the system. Moreover, at least the CA1 region can reconstruct spatial representations when the information transmitted to it from CA3 is drastically degraded (i.e., "pattern completion"). 2) Hippocampal granule cells far outnumber the cortical afferents from which they receive highly convergent input, yet they fire 1:10). Using numerical simulation we will investigate the hypothesis that this organization subserves the orthogonalization or correlated input vectors (c.f. Marr's codon formation). 3). The fascia dentata may also serve in the extraction of novel input features (c.f. the "novelty filter" of Kohonen), an hypothesis based on our observation that net synaptic efficacy increases, yet net excitability is reduced by the acquisition of new information during exploration. To aid in interpreting these phenomena and in the design of future experiments, we will simulate electrically evoked population field potentials in the granule cell network, upon which the theoretical inferences are based, and the possible cellular dynamics that might lead both to these physiological phenomena, and to the proposed novelty extraction effect.