Neuronal Decoding of Decisions in Multidimensional Feature Space Using a Gated Recurrent Variational Autoencoder.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning in Neuroscience

Background:

• Advanced neuroscience tools allow recording from hundreds of neurons during complex cognitive tasks.
• Understanding how the brain processes multidimensional information for decision-making remains a challenge.

Purpose of the Study:

• To develop and apply a novel encode-decode-classify framework to decode decision-making from neural population activity.
• To investigate the neural coding of multidimensional feature learning in the prefrontal cortex and basal ganglia.

Main Methods:

• Utilized a gated recurrent variational autoencoder (VAE) for decoding neural signals.
• Recorded simultaneously from over 300 neuronal channels in monkeys performing a multidimensional task.
• Employed hierarchical stratified sampling and balanced accuracy for model training and evaluation.

Main Results:

• The model achieved high decoding accuracy for decisions in multidimensional feature space.
• Identified distinct neural coding roles: anterior cingulate cortex (ACC) channels collectively encode decision variables, while prefrontal cortex (PFC) channels contribute individually.
• Decoding accuracy was comparable to simpler, lower-dimensional problems.

Conclusions:

• Machine learning frameworks can effectively capture complex spatiotemporal neuronal interactions.
• This approach offers a powerful tool for understanding the neural basis of complex cognitive behaviors.
• The findings shed light on how different brain regions contribute to processing multidimensional information for choice behavior.