Learning, whether in animals or humans, is the process that leads to behaviors better adapted to the environment. This process varies from one individual to another and is typically inferred from their actions during learning. How can we use this individual behavioral data to identify the model that best explains how the individual learns ? I will present two model selection methods: a general hold-out procedure and an AIC-type criterion, both adapted to non-stationary dependent data. I will provide theoretical error bounds for these methods that are close to those of the standard i.i.d. case. Finally, I will give an application of the penalized log-likelihood procedure in ethology for models of learning based on contextual bandits.

Joint work with Louis Köhler, Luc Lehéricy, Giulia Mezzadri and Patricia Reynaud-Bouret