Spontaneous activity is a ubiquitous feature of neural dynamics in the brain, but has no counterpart in artificial information processing machines. It has been suggested that spontaneous activity is not mere noise, but plays active roles in cortical information processing. In my talk, I first present a class of recurrent neural network models that well replicate various statistical features of spontaneous neuronal activity in the hippocampus and neocortex. A key element of these models is long-tailed, typically lognormal, EPSP distributions that were found in both neocortex and hippocampus. These models predict that irregular firing in spontaneous active states is constituted of vast many spike sequences propagating stochastically through local cortical circuits. In the second part, I will use this feature of spontaneous activity to model “preplay” events found in the hippocampus of rodents performing a spatial navigation task. Preplay is a phenomenon in which spontaneous sequences are turned into place-cell sequences during spatial navigation, though some controversial argument exists. Preplay is conceptually intriguing because it suggests that the innate structure of cortical circuits is useful for information coding. Our model suggests that the dendrites of pyramidal cells perform canonical correlation analysis between distal dendritic inputs and proximal dendritic inputs to robustly associate internally driven activity sequences with sensory sequences.
Biography
Dr. Tomoki Fukai is the team leader of the Laboratory for Neural Circuit Theory at Riken Brain Science Institute.