Modeling neural population responses to intracortical microstimulation

Electrical stimulation of a neural population may enable direct experiments relating local population activity with subsequent neural or behavioral processes. However, current stimulation approaches do not attempt precise control of population activity, in part due to the difficulty of characterizing diverse neural responses to varied stimulation patterns. We address this gap by building deep network models that extract neuronal spiking responses through stimulation and predict population responses to arbitrary spatiotemporal stimulation trains. To assess the tractability of training accurate models, we analyze model generalization to novel stimulation patterns and demonstrate that the model can be continuously improved with data aggregated across experimental sessions. Our results show accurate models can be developed within practical experimental budgets, suggesting the feasibility of more precise and flexible stimulation-based population control.