Learning spatio-temporal V1 cells from diverse LGN inputs

Abstract

Since Hubel and Wiesels discovery of simple cells and complex cells in cats primary visual cortex (V1), many experimental studies of V1 cells from animal recordings have shown the spatial and temporal structure of their response properties. Although numerous computational learning models can account for how spatial properties of V1 cells are learnt, how temporal properties emerge through learning is still not well understood. In this study, a learning model based on sparse coding is used to show that spatio-temporal V1 cells, such as biphasic and direction-selective cells, can emerge via synaptic plasticity when diverse spatio-temporal lateral geniculate nucleus (LGN) cells are used as upstream input to V1 cells. We demonstrate that V1 cells with spatial structures and temporal properties (such as a temporal biphasic response and direction selectivity) emerge from a learning process that promotes sparseness while encoding upstream LGN input with spatio-temporal properties. This model provides an explanation for the observed spatio-temporal properties of V1 cells from a learning perspective, enhancing our understanding of how neural circuits learn and process complex visual stimuli.