ESR9 : Reinforcement learning in the hippocampus and medial temporal lobe
Reinforcement learning-based algorithms will be investigated (with DEEPMIND, SISSA), to find out whether model-based or model-free approaches better explain spatial hippocampal activity. Such theoretical ideas will be tested both on neural ensemble data from animals as well as fMRI and MEG data from human subjects navigating virtual environments. The results will be compared with the information analysis developed for rodent data in ESR4. This reinforcement learning-based analysis will extend to inter-area interactions, with the aim of devising a common framework for the interpretation of activity in the extended hippocampal network (hippocampus, parahippocampal regions, retrosplenial cortex, posterior parietal and medial prefrontal) during spatial navigation. The model will also be adapted to the amygdala data from ESR7, and will be compared and combined with statistical physics approaches from ESR10.
Christoffer Gahnstrom was recruited as M-GATE fellow for this project.