Internal KCN: Adaptive Unscented Kalman Filter for Neuronal State and Parameter Estimation

Data assimilation techniques are often used to track, estimate, or correct for estimates of states and parameters of models in a variety of fields. In this work, we investigate how current state estimation techniques, like the robust unscented Kalman filter, apply to single unit neuron models with no prior knowledge about the initial parameters of the neuron model, nor the covariances of the states when applying the Kalman filter. To address this lack of priori knowledge we use an adaptive variant of the robust unscented Kalman filter to adjust the estimates of covariances and compare its performance to related Kalman filters.