Abstract
We present an energy-saving technique for deep spiking neural networks (SNNs) converted from pre-trained analog neural networks (ANNs), named SAPIENS. Deep SNNs have many spiking operations, causing considerable switching power dissipation in the systems that process SNNs. Our study shows that in deep SNNs, only 20% of neurons account for more than 80% of SNN operations, called expensive neurons. Our SAPIENS technique identifies and approximates expensive neurons on-the-fly, improving energy consumption by 50-60% while having minimal impact on SNN accuracy.
| Original language | English |
|---|---|
| Title of host publication | AICAS 2025 - 2025 7th IEEE International Conference on Artificial Intelligence Circuits and Systems, Proceedings |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| ISBN (Electronic) | 9798331524241 |
| DOIs | |
| Publication status | Published - 2025 |
| Event | 7th IEEE International Conference on Artificial Intelligence Circuits and Systems, AICAS 2025 - Bordeaux, France Duration: 28 Apr 2025 → 30 Apr 2025 |
Publication series
| Name | AICAS 2025 - 2025 7th IEEE International Conference on Artificial Intelligence Circuits and Systems, Proceedings |
|---|
Conference
| Conference | 7th IEEE International Conference on Artificial Intelligence Circuits and Systems, AICAS 2025 |
|---|---|
| Country/Territory | France |
| City | Bordeaux |
| Period | 28/04/25 → 30/04/25 |
Bibliographical note
Publisher Copyright:© 2025 IEEE.
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 7 Affordable and Clean Energy
Keywords
- ANN-to-SNN Conversion
- Approximate Computing
- Edge Computing
- Energy Efficiency
- Spiking neural network
Fingerprint
Dive into the research topics of 'SAPIENS: Towards Energy-efficient Deep Spiking Neural Networks via the Approximation of Power-expensive Neurons'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver