neocortex-learning-predictive-error-driven

name	neocortex-learning-predictive-error-driven
description	Neocortex learning framework via error-driven predictive learning using temporal derivatives, corticothalamic circuits, and competitive kinase synaptic plasticity. Implemented in Axon spiking neural simulation framework. Activation: neocortex learning, predictive coding, error-driven learning, temporal derivatives, corticothalamic circuits, kinase plasticity, spiking neurons, Axon framework, competitive learning
metadata	{"arxiv_id":"2606.08720","submitted":"2026-06-07","authors":"Randall C. O'Reilly","tags":["neuroscience","neocortex","learning","predictive-coding","spiking-neural-network","synaptic-plasticity","thalamus","computational-neuroscience"]}
license	Complete terms in LICENSE.txt

Neocortex Learning via Predictive Error-Driven Temporal Derivatives

Context

A sufficient account of how the neocortex learns must meet three criteria: computational (powerful general-purpose learning algorithm), algorithmic (implementable with known neural circuits), and implementational (detailed neurochemical account). Error-driven predictive learning via temporal derivatives, driven by corticothalamic circuits and competitive kinase synaptic plasticity, is the only framework meeting all three criteria.

Core Methodology

Three-Criterion Framework

1. Computational Criterion: Must approximate a powerful, general-purpose learning algorithm known to scale to human-level intelligence
2. Algorithmic Criterion: Must be implementable using known, well-established neural circuits within neocortex and associated brain structures
3. Implementation Criterion: Must provide detailed account of how all algorithmic mechanisms function at neurochemical level

Error-Driven Predictive Learning

1. Temporal Derivative Mechanism: Learning driven by temporal differences in predictions vs outcomes
2. Predictive Coding: Cortex generates predictions about incoming inputs
3. Error Signal Generation: Mismatch between predictions and actual inputs produces error signals
4. Synaptic Plasticity Induction: Errors drive synaptic weight updates via competitive kinase mechanisms

Corticothalamic Circuit Architecture

1. Thalamic Relay: Thalamus acts as relay station for sensory inputs
2. Cortical Feedback: Cortex sends predictive feedback to thalamus
3. Error Detection: Thalamic circuits detect prediction errors by comparing input vs feedback
4. Error Propagation: Errors propagate back to cortex for learning

Competitive Kinase Plasticity

1. Kinase Competition: Multiple kinases compete for synaptic modification control
2. Timing-Dependent Plasticity: Plasticity depends on temporal dynamics of error signals
3. Neurochemical Cascade: Detailed biochemical pathway from error detection to synapse modification
4. Stability Mechanism: Competition ensures stable, non-destructive learning

Implementation in Axon Framework

Spiking Neuron Implementation

1. Axon Framework: Neural simulation framework using spiking neurons
2. Biological Realism: Implements realistic neural dynamics and circuit architecture
3. Circuit Topology: Corticothalamic loops with proper connectivity patterns
4. Temporal Dynamics: Spike-timing-dependent error signal generation

Learning Mechanisms

1. Prediction Generation: Cortical layers generate spike-based predictions
2. Error Computation: Temporal derivative computed from prediction vs actual spikes
3. Plasticity Application: Synaptic weights modified based on error signals
4. Task Learning: Demonstrated across cognitively motivated tasks

Verification Approach

1. Task Performance: Test learning across challenging cognitive tasks
2. Circuit Validation: Verify corticothalamic circuit implementation matches biological data
3. Plasticity Verification: Confirm kinase-based plasticity matches neurochemical evidence
4. Scalability Testing: Demonstrate generalization across task complexity

Key Results

• Implemented in Axon framework using spiking neurons
• Demonstrated learning across wide range of cognitively motivated tasks
• Meets all three criteria: computational, algorithmic, implementational
• Provides complete account from algorithm to neurochemistry

Applications

1. Computational Neuroscience: Unified theory of cortical learning
2. Spiking Neural Networks: Biologically plausible learning rules for SNNs
3. Brain-Computer Interfaces: Understanding cortical plasticity for BCI design
4. Neuromorphic Computing: Implementing predictive learning in neuromorphic hardware
5. Clinical Translation: Understanding learning deficits in neurological disorders

Pitfalls

• Temporal Precision: Error computation requires precise spike timing — ensure sufficient temporal resolution in simulation
• Circuit Complexity: Corticothalamic loops have many subcircuits — validate each component independently
• Kinase Dynamics: Multiple kinase cascades — track competition dynamics carefully to avoid instability
• Prediction Accuracy: Predictions must be sufficiently accurate to generate useful error signals — tune prediction generation mechanism
• Stability Trade-off: Competitive plasticity can suppress learning — balance stability vs plasticity mechanisms

Verification

1. Computational Power: Verify learning algorithm scales to complex tasks (target: human-level performance on standard benchmarks)
2. Circuit Match: Compare implemented corticothalamic circuit with biological data (target: >80% topological similarity)
3. Neurochemical Accuracy: Validate kinase plasticity mechanisms with experimental data
4. Task Generalization: Test across diverse cognitive tasks (pattern recognition, sequence learning, decision making)
5. Stability Analysis: Verify learned representations remain stable over time

Activation Keywords

• neocortex learning
• predictive coding
• error-driven learning
• temporal derivatives
• corticothalamic circuits
• kinase synaptic plasticity
• spiking neurons
• Axon framework
• cortical learning theory
• thalamic feedback
• competitive plasticity