Computational Mechanisms of Learning and Memory

The Stanford Center for Mind Brain and Computation presents:

1:00-1:10      Welcome and Update on the Center for Mind, Brain, and Computation  (Surya Ganguli, Stanford)                                                                         1:10-1:20      Introduction and Overview (Anthony Wagner, Stanford)                                                                                  1:20-2:15      “A feasible detailed probe of the grid cell microcircuit” (Ila Fiete, University of Texas, Austin)        The phenomenology of grid cell population activity has rapidly advanced, but with disparate competing possibilities the circuit mechanisms underlying grid cell activity remain almost entirely unresolved. In this talk, I will propose a strategy that combines existing experimental techniques in a way that promises to bring the mechanistic underpinnings of the grid cell response into considerably sharper focus. The strategy is based on the theoretical insight that small global perturbations of circuit activity will result in characteristic quantal shifts in the spatial tuning relationships between cells, which should be observable from multi- single unit recordings of a small subsample of the population. The predicted shifts differ qualitatively across the candidate recurrent network mechanisms, and also distinguish between recurrent versus feedforward mechanisms. I will discuss how, more generally, the proposed strategy demonstrates that sparse neural recordings coupled with global circuit perturbation can reveal much more about cortical circuit mechanism as it relates to function than can full knowledge of the synaptic connectivity matrix or of network activity.   2:15-3:10      "Entorhinal cortex, acetylcholine and the coding of time and space for memory (Michael Hasselmo, Boston University) Memory function is impaired by lesions of the entorhinal cortex, and also by lesions of the modulatory input from medial septum. My research focuses on understanding circuit mechanisms in the entorhinal cortex that might contribute to memory function.  Neurophysiological data demonstrates cholinergic activation of intrinsic persistent spiking in entorhinal neurons.  This persistent spiking could maintain memory over time in tasks such as delayed matching to sample and trace conditioning and enhance encoding of episodic memory.  These effects might also contribute to coding of time in the phenomenon of “time cells” in the entorhinal cortex and hippocampus.  Cellular mechanisms might also be relevant to the coding of spatial dimensions by grid cells.  Attractor dynamic models of grid cells need a mechanism for spiking interactions to persist over interspike intervals.  The maintenance of attractors could be supported by persistent spiking or rebound spiking in entorhinal neurons.  Loss of cholinergic activation of persistent spiking could underlie the loss of grid cells during inactivation of medial septum, as supported by loss of grid cell firing after cholinergic blockade. These intrinsic properties could also contribute to the phenomenon of theta phase precession in grid cells and theta cycle skipping observed in rodent medial septum and entorhinal cortex.   3:10-3:30      Coffee break                                                                                                                                                 3:30-4:25      “Is the hippocampus an autoassociator for general, non-spatial events?” (Bruce L. McNaughton, University of California, Irvine and The University of Lethbridge) The location where rat hippocampal cells fire in the environment appears to be established primarily the path integration system, although exactly how is not well understood; however, the rates at which cells fire in their place fields appears to be driven by additional cues coming from the environment and/or internal state variables such as working memory and planning of future trajectories.  The external (to the hippocampus) cues drive a process now called ‘rate remapping’, which can rapidly change the rates, but not the locations of place cell firing.  Previously we obtained evidence that the spatial component exhibits evidence of attractor dynamics whereas the rate component does not.  This creates problems for the theory of the hippocampus as a general autoassociator and for the theory that it can preserve in memory unique ‘index codes’ for the attributes of memories stored in distributed neocortical modules.  I will discuss some new evidence and a somewhat more sophisticated conceptual model which seems to rescue the classic theories.   Time permitting, I will also address some current issues surrounding the topic of pattern separation, another element of the classic theory.   4:30-5:00      Panel and audience discussion (moderator: Lisa Giocomo, Stanford)                                                                                           5:00-6:00      Wine and Cheese Reception 2nd floor Harman Terrace, Huang Engineering Building