A generative spike prediction model using behavioral reinforcement for re-establishing neural functional connectivity

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Restoring neural functional connectivity is crucial for treating neurological disorders.
• Traditional spike generation models require downstream recordings, which are unavailable in disease states.
• Existing methods struggle with adaptation and biomimetic encoding.

Purpose of the Study:

• To develop a generative spike model that bypasses the need for downstream recordings.
• To create a reinforcement learning (RL)-based framework for spike train generation.
• To enable biomimetic spike encoding for neural prostheses.

Main Methods:

• A reinforcement learning (RL)-based point process framework was developed.
• The model directly maximizes behavior-level rewards to generate spike trains.
• Upstream neural activity is transformed into behavior-modulated spike patterns.

Main Results:

• The RL-based generative models produced movement-modulated spike patterns similar to healthy subjects.
• The framework demonstrated biomimetic spike encoding capabilities.
• The RL framework outperformed existing methods in generating spike trains.

Conclusions:

• The RL-based spike generation framework offers a promising solution for restoring neural communication.
• This approach shows strong adaptation capabilities across different decoder settings.
• The technology holds potential for advanced neural prostheses and biomimetic cortical stimulation.