A Neural Network Model Inspired by the Prefrontal-Thalamic-Hippocampal Circuits for Navigation under Complex Spatio-temporal Rules

I investigated the functional relationship between the brain regions related to contextdependent memory and aimed to find features of their interactions, particularly information flows among these regions, through brain function modeling using machine learning techniques. I focused on the interaction between the Hippocampus(HPC), Prefrontal cortex(PFC), and nucleus reuniens (Re), which is a component of the thalamus, for the retrieval of contextual information. The modules describing these regions are built using bidirectionally interconnected long-short-term memory (LSTM) units. My results highlight the critical roles of distinct brain modules. These roles and functional interplay enhance the robustness of the system in handling complex cognitive tasks. Furthermore, I simulated a similar network model with reinforcement learning and demonstrated the advantages of the multi-module network structure in rapid adaptation to changes in contextual information over the single-module network structure. In sum, the proposed model inspired by the PFC-ReHPC neural circuit demonstrates robustness and flexibility in learning navigation in complex environments, suggesting the functional benefits of the multi-module network.