Decoding decision-making behavior from sparse neural spiking activity

Decoding animal decision-making behavior from neural spike data emerges as a particularly challenging problem in neuroscience. To address this, we have devised a decision-making decoding model, incorporating a channel attention bi-directional long short-term memory network (CA-BiLSTM), with the aim of effectively parsing sparse neural spike data across multiple brain regions. Notably, the attention mechanism embedded within this model serves the crucial purpose of adeptly localizing neurons integral to decision-making stability within the task at hand. Specifically, when applied to the reproducible electrophysiology dataset from the International Brain Laboratory (IBL), our proposed model has demonstrated a remarkable capacity for accurately forecasting decision-making behavior in mice. Consequently, this investigation furnishes a novel perspective for unraveling the intricacies of neural decision-making mechanisms.