Physics-constrained Inference of Somatic Dynamics from Dendritic Recordings in Weakly Coupled Two-Compartment Neuron Model

This paper addresses the inverse problem of reconstructing somatic membrane potential dynamics from distal dendritic recordings in experimental setups where direct somatic measurements are unavailable. The proposed method employs a physics-informed neural network constrained by a two-compartment neuron model with weak soma-dendrite coupling, where dendritic signals represent filtered and attenuated versions of somatic activity. The approach leverages sparse somatic supervision to guide the inference when somatic data is partially accessible.