Energy optimization induces predictive-coding properties in a multi-compartment spiking neural network model

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Neuroscience

Background:

• Predictive coding is a key theory for brain sensory processing.
• Implementing predictive coding in cortical neuron networks remains challenging.
• Previous work focused on hand-wired circuits or rate-based networks.

Purpose of the Study:

• Investigate if predictive coding emerges from energy optimization in spiking neural networks.
• Explore self-organization principles for cortical connectivity.
• Understand emergent properties in biologically plausible neural models.

Main Methods:

• Developed a multi-compartment spiking neural network model.
• Trained the network with both a task-relevant objective and an energy optimization objective.
• Analyzed network responses to expected and unexpected stimuli.

Main Results:

• The energy-optimized model reconstructed internal representations using top-down expectations.
• Neurons showed distinct responses to expected versus unexpected stimuli.
• Behavior qualitatively matched experimental evidence for predictive coding.

Conclusions:

• Predictive coding-like behavior can emerge from energy optimization principles.
• Energy efficiency may drive self-organization in cortical connectivity.
• Provides a novel perspective on neural implementation of predictive coding.