The neuron as a direct data-driven controller

Area of Science:

• Computational neuroscience
• Neurobiology
• Control theory

Background:

• Existing normative models primarily focus on prediction.
• Gaps in physiological data necessitate new approaches to modeling neuronal function.
• Traditional neuron models (e.g., McCulloch-Pitts-Rosenblatt) are limited in scope.

Purpose of the Study:

• To conceptualize neurons as optimal feedback controllers.
• To extend normative theories of neuronal function beyond prediction.
• To develop a biologically informed model of neuronal control.

Main Methods:

• Utilizing the direct data-driven control (DD-DC) framework.
• Modeling neurons as controllers that steer their environment toward desired states.
• Incorporating synaptic feedback for control effectiveness evaluation.

Main Results:

• The DD-DC neuron model explains spike-timing-dependent plasticity (potentiation-depression shift).
• The model accounts for the duration and adaptive nature of neuronal filters.
• It elucidates spike generation imprecision and brain's operational variability/noise.

Conclusions:

• Neurons function as optimal feedback controllers, not just predictors.
• The DD-DC framework offers a biologically plausible model for neural computation.
• This approach provides a fundamental, modern unit for neural network construction.