FF-INT8: Efficient Forward-Forward DNN Training on Edge Devices with INT8 Precision

2025-06-28

AI Summary

To address the high energy consumption, substantial memory overhead, and insufficient research on low-precision quantized training during backpropagation on edge devices, this paper proposes an INT8-quantized on-device training framework based on the Forward-Forward (FF) algorithm. We innovatively design a layer-wise quantization stabilization strategy and a “look-ahead” mechanism to significantly improve convergence and accuracy stability of FF under low-precision constraints. Furthermore, we integrate activation sharing and memory optimization techniques to reduce system resource utilization. Experiments on the Jetson Orin Nano platform demonstrate that, compared to the baseline, our method achieves a 4.6% speedup in training, an 8.3% reduction in energy consumption, and a 27.0% decrease in memory footprint—while maintaining model accuracy comparable to state-of-the-art methods. This work provides a practical pathway for efficient on-device training in resource-constrained environments.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Backpropagation has been the cornerstone of neural network training for decades, yet its inefficiencies in time and energy consumption limit its suitability for resource-constrained edge devices. While low-precision neural network quantization has been extensively researched to speed up model inference, its application in training has been less explored. Recently, the Forward-Forward (FF) algorithm has emerged as a promising alternative to backpropagation, replacing the backward pass with an additional forward pass. By avoiding the need to store intermediate activations for backpropagation, FF can reduce memory footprint, making it well-suited for embedded devices. This paper presents an INT8 quantized training approach that leverages FF's layer-by-layer strategy to stabilize gradient quantization. Furthermore, we propose a novel "look-ahead" scheme to address limitations of FF and improve model accuracy. Experiments conducted on NVIDIA Jetson Orin Nano board demonstrate 4.6% faster training, 8.3% energy savings, and 27.0% reduction in memory usage, while maintaining competitive accuracy compared to the state-of-the-art.

Problem

• Enables efficient DNN training on edge devices using INT8 precision
• Reduces memory and energy consumption in neural network training
• Improves Forward-Forward algorithm accuracy with a look-ahead scheme

Innovation

• INT8 quantized training with Forward-Forward algorithm
• Layer-by-layer strategy stabilizes gradient quantization
• Look-ahead scheme enhances model accuracy