Fractional-Order Spiking Bayesian Neural Model for Cognitive Computations

Area of Science:

• Computational Neuroscience
• Biophysics
• Cognitive Science

Background:

• Bayesian inference offers a framework for cognition under uncertainty.
• Implementing Bayesian models with biologically plausible neural networks is challenging.
• Existing models lack long-term memory crucial for complex cognitive tasks.

Purpose of the Study:

• To develop a biologically plausible Bayesian neural model for lifespan prediction.
• To integrate fractional-order dynamics into spiking neuron models (Leaky Integrate-and-Fire, Izhikevich).
• To investigate the impact of neural population size on model performance and biological plausibility.

Main Methods:

• Integration of fractional derivatives into Leaky Integrate-and-Fire and Izhikevich neuron models.
• Development of a Bayesian neural model for lifespan prediction using demographic priors.
• Systematic increase in neural population size to assess effects on prediction accuracy and cortical plausibility.
• Comparison of model predictions against human and optimal Bayesian predictions.

Main Results:

• Fractional-order neuron models demonstrated closer alignment with human and optimal Bayesian predictions.
• The large-scale fractional-order Izhikevich model exhibited robust convergence and high cortical plausibility.
• Increased neural population size improved biological plausibility and Bayesian optimality.
• Fractional dynamics enhance representational capacity and introduce necessary long-term memory.

Conclusions:

• Fractal neural dynamics play a key role in probabilistic cognition.
• Biologically inspired spiking neuron models can effectively approximate Bayesian inference.
• This approach bridges theoretical Bayesian models with realistic neural computation, offering pathways for efficient learning, prediction, and adaptation.