Abstract
Recent advances in techniques for monitoring and perturbing neuralpopulations have greatly enhanced our ability to study circuits in the brain.In particular, two-photon holographic optogenetics now enables precisephotostimulation of experimenter-specified groups of individual neurons, whilesimultaneous two-photon calcium imaging enables the measurement of ongoing andinduced activity across the neural population. Despite the enormous space ofpotential photostimulation patterns and the time-consuming nature ofphotostimulation experiments, very little algorithmic work has been done todetermine the most effective photostimulation patterns for identifying theneural population dynamics. Here, we develop methods to efficiently selectwhich neurons to stimulate such that the resulting neural responses will bestinform a dynamical model of the neural population activity. Using neuralpopulation responses to photostimulation in mouse motor cortex, we demonstratethe efficacy of a low-rank linear dynamical systems model, and develop anactive learning procedure which takes advantage of low-rank structure todetermine informative photostimulation patterns. We demonstrate our approach onboth real and synthetic data, obtaining in some cases as much as a two-foldreduction in the amount of data required to reach a given predictive power. Ouractive stimulation design method is based on a novel active learning procedurefor low-rank regression, which may be of independent interest.