A fast and flexible approximation of power-law adaptation for auditory computational models

Abstract

1.AbstractPower-law adaptation is a form of neural adaptation that has been shown to provide a better description of auditory-nerve adaptation dynamics as compared to simpler exponential-adaptation processes. However, the computational costs associated with power-law adaptation are high and, problematically, grow superlinearly with the number of samples in the simulation. This cost limits the applicability of power-law adaptation in simulations of responses to relatively long stimuli, such as speech, or in simulations for which high sampling rates are needed. Here, we present a simple approximation to power-law adaptation based on a parallel set of exponential-adaptation processes with different time constants, demonstrate that the approximation improves on an existing approximation provided in the literature, and provide updates to a popular phenomenological model of the auditory periphery that implements the new approximation.

A Computational Model of Auditory Chirp-Velocity Sensitivity and Amplitude-Modulation Tuning in Inferior Colliculus Neurons

“…1). Inputs to the octopus-cell stage were provided by a version of the Zilany et al (2014) AN model that was modified to include gain control via the medial olivocochlear (MOC) efferent (Farhadi et al, 2023) and an improved approximation to the power-law synapse model (Guest and Carney, 2023). The efferent feedback in the AN model affected responses to sounds with modulated envelopes, including the aperiodic random chirp stimulus used here to characterize model neurons' chirp-velocity sensitivity.…”