Adaptive sampling for a population of neurons

Adaptive sampling methods in neuroscience have primarily focused on maximizing the firing rate of a single recorded neuron. When recording from two or more neurons, it is usually not possible to find a single stimulus that maximizes the firing rates of all neurons. This motivates an objective function that takes into account the recorded population of neurons together. We propose ``Adept,'' an adaptive sampling method that can optimize population objective functions. In simulated experiments, we first confirmed that population objective functions elicited more varied stimulus responses than those of single-neuron objective functions. Then, we tested Adept in a closed-loop electrophysiological experiment in which population activity was recorded from macaque V4, a cortical area known for mid-level visual processing. Adept uses the outputs of a deep convolutional neural network model as feature embeddings to predict neural responses. Adept elicited mean stimulus responses 20\% larger than those for randomly-chosen natural images, as well as a larger scatter of stimulus responses. Such adaptive sampling methods can enable new scientific discoveries when recording from a population of neurons with heterogeneous response properties.