Jonathan Pillow

Wednesday - 10 October 2018

Time: 4.00pm

Ground Floor Seminar Room

25 Howland Street, London, W1T 4JG

Efficient methods for learning and control of animal behavior

The process of learning new behaviors is of great interest in neuroscience and artificial intelligence. However, most standard analyses of animal training data either treat behavior as fixed, or track only coarse performance statistics (e.g., accuracy and bias), providing limited insight into behavioral strategies that evolve over the course of learning. To overcome these limitations, we propose a dynamic psychophysical model that efficiently tracks trial-to-trial changes in behavior over the course of training. In this talk, I will describe new work based on a dynamic logistic regression model, which captures the dynamic dependencies of behavior on stimuli and common task-irrelevant variables including choice history, sensory history, reward history, and choice bias. We have applied our method to psychophysical data from both human subjects and rats learning a delayed sensory discrimination task, and can successfully track the dynamics of psychophysical weights during training, capturing day-to-day and trial-to-trial fluctuations that underlie changes in performance, choice bias, and dependencies on task history. We leverage the model's flexibility model to investigate why rats frequently make mistakes on easy trials, demonstrating that so-called "lapses" often arise from sub-optimal weighting of task covariates. Finally, I will describe recent work on adaptive optimal training, which combines ideas from reinforcement learning and adaptive experimental design to formulate methods for inferring animal learning rules from behavior, and using these rules to speed up animal training.