Convolutional method for data assimilation An improved method on neuronal electrophysiological data

Abstract

We present a convolution-based data assimilation method tailored to neuronal electrophysiology, addressing the limitations of traditional value-based synchronization approaches. While conventional methods rely on nudging terms and pointwise deviation metrics, they often fail to account for spike timing precision, a key feature in neural signals. Our approach applies a Gaussian convolution to both measured data and model estimates, enabling a cost function that evaluates both amplitude and timing alignment via spike overlap. This formulation remains compatible with gradient-based optimization. Through twin experiments and real hippocampal neuron recordings, we demonstrate improved parameter estimation and prediction quality, particularly in capturing sharp, time-sensitive dynamics.