Decades of research in the laboratory setting has led to big advances in our understanding of how animals learn, and the neural circuits involved. Surprisingly, most of such research involves behavioral tasks that essentially test memorization, using the same stimuli, task rules or motor outputs in the testing phase as those used in the training. This contrasts machine learning systems, where generalization is considered a fundamental test of learning, and emphasizes performance evaluation with new in-distribution or out-of-distribution examples. We are developing behavioral tasks in mice to understand the brain basis of generalization. I will present a concrete example on olfactory learning in mice that involves detection of relevant odorants embedded within complex mixtures of distractors.
These experiments suggest that mice have an inductive bias towards generalization.