Bayesian model selection and minimum description length estimation of auditory-nerve discharge rates.

Area of Science:

• Auditory Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Auditory-nerve fiber discharges are modeled as self-exciting point processes.
• Existing maximum-likelihood methods face over-parameterization issues.
• Accurate modeling is crucial for understanding auditory processing.

Purpose of the Study:

• To develop a novel procedure for estimating stimulus-related and refractory-related functions in auditory-nerve fiber discharge models.
• To address and solve the over-parametrization problem inherent in previous estimation methods.
• To compare the performance of the new procedure against existing maximum-likelihood algorithms.

Main Methods:

• Utilized a Bayes criterion for model complexity selection and parameter estimation.
• Developed a procedure asymptotically related to Rissanen's minimum description length (MDL) criterion.
• Compared the performance of the MDL procedure with traditional maximum-likelihood algorithms.

Main Results:

• The new Bayes criterion-based procedure effectively solves the over-parametrization problem.
• The MDL procedure demonstrates superior performance compared to previous maximum-likelihood methods.
• The proposed method allows for simultaneous estimation of stimulus and recovery properties.

Conclusions:

• The MDL procedure offers a robust and accurate method for modeling auditory-nerve fiber discharges.
• This approach improves the estimation of both stimulus encoding and neural recovery dynamics.
• Recommends the adoption of the MDL procedure for auditory nerve modeling research.