A hybrid generative and predictive model of the motor cortex

Area of Science:

• Computational Neuroscience
• Robotics
• Cognitive Science

Background:

• The motor cortex plays a crucial role in planning and executing movements.
• Understanding the neural mechanisms underlying motor control is essential for advancing artificial intelligence and treating neurological disorders.

Purpose of the Study:

• To develop a hybrid generative and predictive computational model of the motor cortex.
• To investigate the model's ability to perform learned action sequences and mental simulations.
• To explore the potential role of the motor cortex in functions previously attributed to the basal ganglia, such as reinforcement learning.

Main Methods:

• A hybrid model combining generative and predictive components was developed.
• The generative model utilizes hierarchically directed cortico-cortical connections for unsupervised learning of topographic and sparse representations.
• The predictive model employs lateral cortical connections as a hetero-associator attractor network for predicting future network states.

Main Results:

• The generative model successfully mapped sensory input to motor actions, enabling the execution of learned action sequences.
• The predictive model demonstrated the capacity for mental simulation of perception- and action sequences.
• Model performance was validated using a visually guided robot docking maneuver.

Conclusions:

• The proposed hybrid model offers a new framework for understanding motor cortex function.
• The motor cortex may integrate functions of reinforcement learning, potentially involving mirror neurons and imitation.
• This research provides a computational basis for exploring motor learning and cognitive processes.