Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications

2024-09-05

AI Summary

To address high inference complexity, reliance on auxiliary quantization training, and difficulties in plug-and-play deployment of pretrained artificial neural networks (ANNs) in ANN-to-spiking neural network (SNN) conversion, this paper proposes an efficient conversion framework with inference-scale computational complexity. Our method eliminates the need for retraining by introducing (1) a novel local threshold balancing algorithm coupled with channel-wise threshold scaling for training-free quantization, and (2) a latency-aware evaluation strategy to effectively compensate for spike propagation delays. The framework enables direct deployment of standard pretrained ANNs without architectural modification. It achieves state-of-the-art accuracy across diverse vision tasks—including image classification, semantic segmentation, object detection, and video classification—while significantly reducing power consumption compared to their ANN counterparts. Accuracy degradation is negligible, demonstrating a compelling trade-off between high-speed inference and ultra-low energy consumption.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Learning & Optimization for CV
• Machine Learning: Learning on the Edge & Model Compression

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Spiking Neural Networks (SNNs) have emerged as a promising substitute for Artificial Neural Networks (ANNs) due to their advantages of fast inference and low power consumption. However, the lack of efficient training algorithms has hindered their widespread adoption. Even efficient ANN-SNN conversion methods necessitate quantized training of ANNs to enhance the effectiveness of the conversion, incurring additional training costs. To address these challenges, we propose an efficient ANN-SNN conversion framework with only inference scale complexity. The conversion framework includes a local threshold balancing algorithm, which enables efficient calculation of the optimal thresholds and fine-grained adjustment of the threshold value by channel-wise scaling. We also introduce an effective delayed evaluation strategy to mitigate the influence of the spike propagation delays. We demonstrate the scalability of our framework in typical computer vision tasks: image classification, semantic segmentation, object detection, and video classification. Our algorithm outperforms existing methods, highlighting its practical applicability and efficiency. Moreover, we have evaluated the energy consumption of the converted SNNs, demonstrating their superior low-power advantage compared to conventional ANNs. This approach simplifies the deployment of SNNs by leveraging open-source pre-trained ANN models, enabling fast, low-power inference with negligible performance reduction. Code is available at https://github.com/putshua/Inference-scale-ANN-SNN.

Problem

• Efficient ANN-SNN conversion for low-power applications
• Addressing lack of efficient training algorithms in SNNs
• Mitigating spike propagation delays in SNN inference

Innovation

• Local threshold balancing algorithm for ANN-SNN conversion
• Delayed evaluation strategy to reduce spike delays
• Energy-efficient SNNs with minimal performance loss