Using economic value signals from primate prefrontal cortex in neuro-engineering applications.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Brain-machine interfaces (BMIs) traditionally use motor/sensory signals.
• Abstract cognitive signals offer untapped potential for neuro-engineering.
• Economic value is a key cognitive construct for decision-making.

Purpose of the Study:

• To explore the use of neural signals related to economic value in a BMI context.
• To develop deep learning-based neural decoders for predicting choices in value-based decision-making tasks.
• To implement adaptive decoders for multi-step decisions using reinforcement learning.

Main Methods:

• Collected multivariate time series data from the orbitofrontal cortex in non-human primates.
• Developed deep learning neural decoders to predict choices.
• Utilized a reinforcement learning-based training approach for adaptive decoders.

Main Results:

• Achieved >70% average accuracy in predicting monkey choices using subjective value signals.
• Demonstrated above-chance accuracy even with objectively equal choice options.
• Showed decoder architecture can execute choice-related actions and action sequences.
• Developed a neural forecasting model predicting choices ~300 ms sooner.

Conclusions:

• User preference-informed neuro-engineering devices leveraging cognitive signals are feasible.
• Combining abstract cognitive signals with motor/sensory data may enhance accuracy.
• Future systems may require confidence measurement for minimal user input scenarios.