Hybrid ANN-SNN Pipeline with Local Plasticity

2026-06-19 · Source: cs.NE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, long

Summary

A hybrid Artificial Neural Network-Spiking Neural Network (ANN-SNN) pipeline is proposed, integrating a pretrained EfficientNet-B3 encoder with a CoLaNET spiking classifier. This architecture converts the EfficientNet-B3 encoder's activations into spike trains using rate-coding, then trains the CoLaNET classifier with local, biologically inspired learning rules, avoiding end-to-end gradient propagation. Evaluated on a 64-class ImageNet subset comprising 35,179 images, the system achieved 99.09% classification accuracy. The EfficientNet-B3 encoder, with 10.8M parameters, extracts 1536-dimensional feature vectors. Hyperparameter optimization for the SNN classifier, which uses an ensemble of 15 CoLaNET instances, was performed via a genetic algorithm over three days on a GPU cluster. This online, single-pass learning approach demonstrates performance on par with conventional deep networks.

Key takeaway

For Machine Learning Engineers designing energy-efficient, continuously learning systems, you should consider hybrid ANN-SNN architectures. This approach allows you to combine powerful pretrained ANN encoders for feature extraction with local plasticity SNNs for efficient, online decision-making. While the ANN encoder remains a power bottleneck, this strategy offers a practical path to neuromorphic hardware, especially for on-device continual learning applications. Future work should focus on converting the ANN encoder to a fully spiking model.

Key insights

Hybrid ANN-SNN pipelines achieve high accuracy with local, biologically plausible learning by utilizing pretrained ANN encoders.

Principles

• Decouple feature extraction from decision-making.
• Local plasticity enables online, single-pass learning.
• Ensembles improve weak individual classifiers.

Method

The method involves converting EfficientNet-B3 activations to spike trains via rate-coding, then training a CoLaNET spiking classifier using local plasticity rules and a genetic algorithm for hyperparameter optimization.

In practice

• Use pretrained ANNs for robust feature extraction.
• Implement SNNs with local rules for online learning.
• Employ genetic algorithms for SNN hyperparameter tuning.

Topics

• Hybrid Neural Networks
• Spiking Neural Networks
• Local Plasticity
• EfficientNet-B3
• CoLaNET
• Neuromorphic Hardware

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Hardware Engineer

Related on AIssential

• Hybrid ANN-SNN Pipeline with Local Plasticity — A hybrid ANN-SNN pipeline combines pretrained ANN embeddings and local SNN plasticity for high accuracy, bypassing end-to-end gradient training.
• MPCS: Neuroplastic Continual Learning via Multi-Component Plasticity and Topology-Aware EWC — Integrating multiple plasticity mechanisms and using Pareto analysis can optimize continual learning systems.
• Structured Representation Learning with Locally Linear Embeddings and Adaptive Feature Fusion — A neuroscientifically inspired RL framework disentangles dynamics and reward features using LLEs and adaptive fusion for improved learning.
• Memory-Augmented Spiking Networks: Synergistic Integration of Complementary Mechanisms for Neuromorphic Vision — Architectural balance of memory mechanisms in SNNs yields synergistic performance gains over individual optimizations.
• Asymmetric-Loss-Guided Hybrid CNN-BiLSTM-Attention Model for Industrial RUL Prediction with Interpretable Failure Heatmaps — A hybrid CNN-BiLSTM-Attention model with asymmetric loss improves turbofan RUL prediction and interpretability.
• Self-supervised local learning rules learn the hidden hierarchical structure of high-dimensional data — Self-supervised local learning rules can uncover hidden hierarchical data structures with biological plausibility.

Open in AIssential →

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