A tutorial on the free-energy framework for modelling perception and learning.

Area of Science:

• Computational neuroscience
• Cognitive science
• Machine learning

Background:

• Predictive coding models, like Rao and Ballard's, explain sensory inference.
• The sensory cortex infers stimulus features from noisy neural inputs.
• Biological plausibility of these models is a key consideration.

Purpose of the Study:

• To provide an accessible tutorial on Friston's free-energy framework.
• To extend the predictive coding model for perception.
• To detail the implementation of inference and learning in neural networks.

Main Methods:

• Step-by-step derivations of the free-energy model.
• Illustrative examples for clarity.
• Discussion of biological neural circuit implementation.

Main Results:

• The free-energy framework offers a unified approach to perception and learning.
• Models demonstrate inference via simple computational elements, suggesting neural network implementation.
• Hebbian learning rules are used for synaptic plasticity, updating feature parameters and uncertainty.

Conclusions:

• The free-energy framework provides a powerful, biologically plausible model for perception.
• It unifies inference and learning mechanisms within neural systems.
• An extended model highlights simplified neuronal summation and synaptic plasticity rules.