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

TL;DR

This paper introduces an INT8 quantized training method for the Forward-Forward neural network algorithm, enabling efficient, low-memory, and energy-saving training suitable for edge devices, with improved stability and accuracy.

Contribution

It proposes a novel INT8 quantization approach combined with a look-ahead scheme for the Forward-Forward algorithm, enhancing training efficiency and stability on resource-constrained devices.

Findings

4.6% faster training on NVIDIA Jetson Orin Nano

8.3% energy savings during training

27.0% reduction in memory usage

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.