What question did this study set out to answer?

This research aims to develop a formula for effectively converting ANNs to SNNs with no accuracy loss.

synapse

⌘+K

synapse

⌘+K

February 5, 2026Open Access

Activation-Scaled ANN-to-SNN Conversion: A Universal Threshold Formula for Lossless Neural Network Transformation

Key Points

This research aims to develop a formula for effectively converting ANNs to SNNs with no accuracy loss.
Introduced a universal threshold formula: θ = 2.0 × max(activation)
Tested on various neural network architectures like MLP, CNN, and ResNet
Used a hybrid readout strategy with specified spike rates
Conducted the process without additional training, using pure weight copying
Achieved 100% accuracy preservation during conversion
Utilized a hybrid spike rate composition of 70% spikes and 30% membrane potential

Abstract

I present a simple yet effective formula for converting Artificial Neural Networks (ANNs) to Spiking Neural Networks (SNNs) with minimal accuracy loss. Key Findings:- Universal threshold formula: θ = 2.0 × max(activation)- Tested on MLP, CNN, ResNet architectures- 100% accuracy preservation achieved- Hybrid readout: 70% spike rate + 30% membrane potential- No training required - pure weight copy This work provides a practical, training-free conversion method for deploying neural networks on neuromorphic hardware. Code: https://github.com/hafufu-stack/temporal-coding-simulation

Activation-Scaled ANN-to-SNN Conversion: A Universal Threshold Formula for Lossless Neural Network Transformation

Key Points

Abstract

Cite This Study