Modelling neural coding in the auditory midbrain with high resolution and accuracy

Area of Science:

• Neuroscience
• Computational Auditory Processing
• Artificial Intelligence

Background:

• Models of the cochlea are advanced, but computational models of the auditory brain, specifically the inferior colliculus, lag significantly in performance and application.
• Existing models struggle with the statistical structure of neural responses, non-stationarity, and generalizing across brains.

Purpose of the Study:

• To present ICNet, a convolutional encoder-decoder model designed to simulate neural coding in the inferior colliculus.
• To address key challenges in sensory system modeling: capturing neural response patterns, handling non-stationarity, and extracting shared processing features.

Main Methods:

• Developed ICNet using large-scale intracranial recordings from anaesthetized gerbils.
• Employed a convolutional encoder-decoder architecture to model neural coding.
• Focused on statistical structure, physiological non-stationarity, and cross-brain feature extraction.

Main Results:

• ICNet achieved highly accurate simulations of multi-unit neural responses to diverse complex sounds, including speech.
• The model successfully reproduced key neurophysiological phenomena like forward masking and dynamic range adaptation.
• Demonstrated the ability to simulate thousands of neural units and provide compact representations of auditory processing.

Conclusions:

• ICNet represents a significant advancement in auditory brain modeling, offering high fidelity and broad applicability.
• The model can facilitate hearing research, audio technology development, and serve as a foundation for higher-level auditory processing models.
• ICNet's ability to simulate neural activity and represent auditory processing dynamics opens new avenues for understanding and engineering auditory systems.