Model Predictive Control on the Neural Manifold.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Control Theory

Background:

• Neural manifolds offer a framework for understanding neural population activity but lack dynamic control methods.
• Current tools for analyzing neural manifolds are often correlational, limiting insight into circuit dynamics.
• Precise control over latent neural activity is crucial for investigating neural manifold structure and function.

Purpose of the Study:

• To simulate and evaluate methods for controlling latent dynamics within a neural population.
• To compare the efficacy of proportional-integral-derivative (PID) control and model predictive control (MPC) for neural manifold control.
• To establish a framework for experimentally testing causal relationships between neural manifold dynamics and external stimuli.

Main Methods:

• Utilized a spiking neural network (SNN) to model neural circuit dynamics.
• Simulated closed-loop, dynamically generated sensory inputs to control latent activity.
• Evaluated PID and MPC controllers on trajectory-following tasks under partial observability and noise.

Main Results:

• Both PID and MPC controllers showed ability to control latent dynamics on the neural manifold.
• Model predictive control (MPC) consistently achieved more accurate control compared to PID.
• MPC required less hyperparameter tuning and demonstrated robustness in challenging conditions.

Conclusions:

• Model predictive control (MPC) can be effectively applied to neural manifolds using data-driven dynamics models.
• This simulation provides a framework for future experimental validation of control strategies in neural circuits.
• Demonstrates the potential of control theory to uncover causal links between neural dynamics and behavior.