Innovate closed-loop brain stimulation technology using Machine Learning

Brain stimulation has the potential to create desired neural population activity states. However, it is challenging to search the large space of stimulation parameters, for example, selecting which subset of electrodes to be used for stimulation. To address this challenge, we develop MiSO (MicroStimulation Optimization), a closed-loop stimulation framework to drive neural population activity toward specified states by optimizing over a large stimulation parameter space. MiSO consists of three key components: 1) a neural activity alignment method to merge stimulation-response samples across sessions, 2) a statistical model trained on the merged samples to predict the brain’s response to untested stimulation parameter configurations, and 3) an online optimization algorithm to adaptively update the stimulation parameter configuration based on the model’s predictions. MiSO increases the clinical viability of neuromodulation technologies by enabling the use of many-fold larger stimulation parameter spaces.

Achieving a goal in natural environments often requires a sequence of decisions, each bringing us closer to the final outcome. In sequential decisions, previous choices could facilitate or hinder future ones. Therefore, to optimize the choice sequence, it is required to adopt a prospective decision strategy that incorporates the interactions between the current and future choices. Such decision strategies have been reported in humans, but little is known about prospective planning in animal models such as macaque monkeys. Here, we develop a novel maze task where monkeys move a token from a starting position on one side of the screen to a goal on the other side, avoiding three immobile obstacles. The chosen path around the immediate wall puts the token in a position that could facilitate or hinder passing the subsequent wall(s). Therefore, the task examined the monkey’s ability to plan prospectively. We found that monkeys take forthcoming obstacles into account, instead of just choosing the easiest path around the nearest obstacle, and that their planning ”depth” exceeds three future obstacles (the maximum tested by our task). They apply this prospective strategy with little training, suggesting that prospective planning is part of monkeys’ cognitive repertoire. In line with this finding, their strategy generalizes to novel maze patterns that were not shown during the training. Our task offers an opportunity to study the neural mechanisms of prospective planning with monkeys without the need for extensive training typically required when studying higher cognition.