How do neurons maintain reliable communication in the face of noise? This paper introduces a simplified model of spike generation that explains the phenomenon of long-term spike train regularization, where the timing of neural spikes becomes remarkably consistent over time. This regularization is characterized by negative correlations in the interspike interval (ISI) sequence, leading to a variance in the kth-order interval distribution that is significantly smaller than expected for uncorrelated ISIs. Focusing on neural computation, the authors present a linear adaptive threshold model that incorporates a dynamic spike threshold, transiently elevated following a spike, to capture these effects. The model, inspired by electrosensory afferent dynamics, demonstrates that refractory effects associated with the dynamic threshold lead to long-term spike train regularization, enhancing the detectability of weak signals encoded in noisy spike trains. The properties of the **linear adaptive threshold model** can lead to dramatic improvement in the detectability of weak signals encoded in the spike train data. While motivated by electrosensory afferent nerve fibers, the authors suggest that such regularizing effects may play a crucial role in various neural systems requiring reliable detection of weak signals. The linear adaptive threshold model offers a valuable tool for modeling neuronal systems with specific ISI correlation structures.
This article, published in _Neural Computation_, is highly relevant to the journal's focus on computational and theoretical neuroscience. By presenting a simplified model of spike train regularization, the paper directly addresses core themes in neural coding and signal processing within the nervous system.