RECAP: Local Hebbian Prototype Learning as a Self-Organizing Readout for Reservoir Dynamics

2026-03-10 · Source: cs.NE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Advanced, extended

Summary

RECAP (Reservoir Computing with HEbbian Co-Activation Proto-types) is a bio-inspired learning strategy for robust image classification that couples untrained reservoir dynamics with a self-organizing Hebbian prototype readout. It avoids error backpropagation and is compatible with online prototype updates. The method discretizes time-averaged reservoir responses into activation levels, constructs a co-activation mask over reservoir unit pairs, and incrementally updates class-wise prototype matrices via a Hebbian-like potentiation–decay rule. Inference is performed by overlap-based prototype matching. Evaluated on MNIST-C, RECAP achieved a Relative mCE of 34.1% across 15 corruption types at five severity levels, significantly outperforming MLP (52.1%), ESN-Ridge (55.0%), ResNet-18, and AlexNet, despite being trained only on clean samples.

Key takeaway

For AI scientists and research scientists developing robust vision systems, RECAP demonstrates that significant corruption robustness can be achieved through a self-organizing, Hebbian-inspired readout without training on corrupted data or using backpropagation. You should consider exploring local plasticity rules and discrete relational representations as alternatives to end-to-end gradient optimization, especially when aiming for zero-shot robustness and online adaptability in constrained environments.

Key insights

Hebbian co-activation prototypes in reservoir computing enable robust image classification without backpropagation.

Principles

• Robustness can emerge from readout representation, not just corruption-specific training.
• Discretization stabilizes representations against small amplitude perturbations.
• Relational masks emphasize structural patterns over exact feature magnitudes.

Method

RECAP uses an untrained Echo State Network to generate high-dimensional states, which are then discretized into co-activation masks. Class-wise prototypes are updated via a Hebbian potentiation-decay rule, and inference uses overlap matching.

In practice

• Use untrained recurrent networks for feature generation.
• Employ discrete activation levels to enhance perturbation tolerance.
• Implement Hebbian-like rules for local, online prototype updates.

Topics

• Reservoir Computing
• Hebbian Learning
• Robustness
• Bio-inspired AI
• Image Classification

Best for: AI Scientist, Research Scientist, AI Researcher, AI Engineer, Machine Learning Engineer

Related on AIssential

• Benchmarking local Hebbian learning rules for memory storage and prototype extraction — Bayesian-Hebbian learning rules offer superior capacity for associative memory and prototype extraction compared to other Hebbian rules.
• Reservoir computing experiment - a Liquid State Machine with simulated biological constraints (hormones, pain, plasticity) — Simulating biological constraints in reservoir computing creates distinct system dynamics.
• Incremental learning for audio classification with Hebbian Deep Neural Networks — Hebbian learning with kernel plasticity improves incremental audio classification by balancing new learning and knowledge retention.
• Hebbian architecture AI model [R] — A Hebbian AI model demonstrates emergent learning and resilience without backpropagation, using minimal parameters.
• Computer Vision: Chapter 1 (Traditional Chinese) — CNNs leverage hierarchical feature extraction and convolution to enable machines to understand visual data, evolving from biological inspiration and early neural networks.
• Self-supervised local learning rules learn the hidden hierarchical structure of high-dimensional data — Self-supervised local learning rules can effectively uncover hierarchical data structures, aligning with biological plasticity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.NE updates on arXiv.org.