Deep neural networks explain spiking activity in auditory cortex.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Artificial neural networks (ANNs) excel at predicting neural responses in primate visual and auditory cortex for static stimuli.
• The predictive power of ANNs for fine-timescale spiking activity, crucial for audition, remains largely unexplored.

Purpose of the Study:

• To investigate if ANNs trained on speech audio can predict spiking activity in the auditory cortex at fine time scales.
• To compare the predictive performance of trained ANNs against traditional spectrotemporal receptive fields and untrained networks.

Main Methods:

• Utilized ANNs trained on speech audio datasets.
• Performed acute multi-electrode recordings from the auditory cortex of squirrel monkeys.
• Analyzed spike counts of multi-units in response to speech and monkey vocalizations using varying time bin widths (≤50 ms).

Main Results:

• Trained ANNs successfully predicted multi-unit spike counts in the auditory cortex at time scales of 50 ms and below.
• ANNs explained significantly more explainable neural variance compared to traditional spectrotemporal receptive fields and untrained networks.
• Deeper ANN layers showed better predictability for non-primary neurons, though with considerable neuron-specific variation.

Conclusions:

• ANNs trained on speech provide a powerful tool for predicting auditory cortical activity at fine time scales.
• This approach surpasses traditional methods in explaining neural variance, offering new insights into auditory processing.
• The findings highlight the potential of ANNs for understanding neural coding in complex sensory systems.