The Hamiltonian Brain: Efficient Probabilistic Inference with Excitatory-Inhibitory Neural Circuit Dynamics

Area of Science:

• Computational neuroscience
• Systems neuroscience

Background:

• Probabilistic inference is a key framework for understanding brain function.
• Cortical responses exhibit oscillations and transients, which are challenging to model with existing probabilistic inference frameworks.

Purpose of the Study:

• To demonstrate that neural oscillations and transients are inherent features of probabilistic inference algorithms used by the cortex.
• To propose Hamiltonian Monte Carlo (HMC) as a model for cortical probabilistic inference.

Main Methods:

• Developed an excitatory-inhibitory neural network model of primary visual cortex.
• Implemented Hamiltonian Monte Carlo (HMC) inference within the model.
• Analyzed the emergent network dynamics, including oscillations and transients, in response to varying stimulus contrast.

Main Results:

• The HMC-based model naturally produced oscillations and transients, consistent with observed cortical activity.
• Oscillations significantly accelerated the inference process, improving efficiency by an order of magnitude.
• Model dynamics, including oscillation frequency and transient magnitude, scaled with stimulus contrast.
• Balanced excitation and inhibition, with inhibition lagging excitation, were observed.

Conclusions:

• Neural oscillations and transients are functional components of efficient probabilistic inference in the brain.
• Hamiltonian Monte Carlo (HMC) provides a viable algorithmic framework for explaining these dynamics in excitatory-inhibitory cortical circuits.
• The separation of excitatory and inhibitory populations and their temporal dynamics are crucial for efficient neural computation.