Group-Patch Based Classification and Asymptotic Predicting Imbalanced Neuron Spikes

The cerebral cortex is connected to sub-cortical structures through multiple hierarchically organized descending and ascending pathways. Cortical representations of subcortical neural activity reflect embedded complex spatiotemporal dynamics. Past studies extensively examined the molecular, cellular and circuit properties of intracortical and cortico-subcortical pathways. It remains poorly understood how the spiking activity of a cortical or subcortical neuron is associated with cortex-wide network dynamics. In this study, we use simultaneously recorded mesoscale calcium imaging and behavior video as multi-modality for predicting neuron spiking on awake mice. A novel group patch wise classification and asymptotic model is proposed to address the ultra imbalanced spikes prediction problem. Experiments demonstrate the efficacy of the proposed classification and asymptotic predicting strategies. Our empirical results reveal that the spike activities of neurons were associated with distinct cortical calcium signals and behavior information,which was observed that ongoing neural cortical activity encodes a high-dimensional latent signal of behavior and sensory.