O'Reilly's Neocortex Learning Theory: Predictive Error via Corticothalamic Circuits

Randall C. O'Reilly (University of Colorado) presents a comprehensive theory of how the neocortex learns, satisfying three essential criteria. First, it approximates a powerful, general-purpose learning algorithm known to scale to human-level intelligence. Second, it is implementable using well-established neural circuits within the neocortex and associated structures. Third, it provides a detailed neurochemical account of all algorithmic mechanisms. The proposed solution is error-driven predictive learning via temporal derivatives, driven by corticothalamic circuits, based on competitive kinase synaptic plasticity induction.

O'Reilly has implemented this theory in the Axon neural simulation framework using spiking neurons and demonstrated learning across a range of challenging cognitively motivated tasks. The framework unifies bottom-up and top-down signaling, explaining how the brain predicts sensory input and adjusts synaptic weights through precise timing of neural activity. This work has immediate implications for AI: it suggests that architectures incorporating temporal prediction and corticothalamic-like feedback loops could achieve more human-like learning efficiency. The paper is available on arXiv (2606.08720).