Predictive Ensemble Decoding of Acoustical Features Explains Context-Dependent Receptive Fields.

Area of Science:

• Auditory Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Auditory neuroscience traditionally uses linear encoding models to understand how neurons represent sound features.
• Neural responses in the auditory cortex are complex, dynamic, and nonlinear, posing challenges for existing models.
• The influence of network activity on neural responses necessitates exploring alternative modeling approaches.

Purpose of the Study:

• To investigate an alternative approach to auditory neuroscience: decoding the stimulus from neural responses.
• To develop and validate a Bayesian normative decoding model for auditory cortical neurons.
• To compare the predictive power of decoding models against traditional encoding models.

Main Methods:

• Utilized a Bayesian normative approach to model neural responses.
• Compared model predictions with electrophysiological recordings from the primary auditory cortex of ferrets.
• Analyzed neuronal selectivity and response dynamics in relation to stimulus features.

Main Results:

• Decoding filters derived from neural activity align with statistical properties of speech sounds.
• The Bayesian decoding model demonstrates superior performance in capturing response dynamics compared to linear encoding models.
• The decoding model accurately accounts for stimulus representation accuracy and nonlinearities, outperforming encoding models.

Conclusions:

• Neuronal responses are shaped by a process of "explaining away," involving divisive competition between interpretations of auditory input.
• A decoding framework provides a more effective approach for understanding auditory cortical function and neural representations of sound.
• This study highlights the importance of considering network interactions and Bayesian principles in auditory neuroscience.