ssProp: Energy-Efficient Training for Convolutional Neural Networks with Scheduled Sparse Back Propagation

TL;DR

ssProp introduces a novel energy-efficient training method for CNNs that employs scheduled sparse backpropagation, significantly reducing computation and energy consumption while maintaining or improving model performance across various tasks.

Contribution

The paper proposes a general, energy-efficient convolution module with channel-wise sparsity and gradient schedulers, enabling sparse backpropagation across diverse architectures and tasks.

Findings

Reduces 40% of computations during training.

Potentially improves model performance by mitigating over-fitting.

Compatible with various datasets and deep learning architectures.

Abstract

Recently, deep learning has made remarkable strides, especially with generative modeling, such as large language models and probabilistic diffusion models. However, training these models often involves significant computational resources, requiring billions of petaFLOPs. This high resource consumption results in substantial energy usage and a large carbon footprint, raising critical environmental concerns. Back-propagation (BP) is a major source of computational expense during training deep learning models. To advance research on energy-efficient training and allow for sparse learning on any machine and device, we propose a general, energy-efficient convolution module that can be seamlessly integrated into any deep learning architecture. Specifically, we introduce channel-wise sparsity with additional gradient selection schedulers during backward based on the assumption that BP is often dense and inefficient, which can lead to over-fitting and high computational consumption. Our experiments demonstrate that our approach reduces 40\% computations while potentially improving model performance, validated on image classification and generation tasks. This reduction can lead to significant energy savings and a lower carbon footprint during the research and development phases of large-scale AI systems. Additionally, our method mitigates over-fitting in a manner distinct from Dropout, allowing it to be combined with Dropout to further enhance model performance and reduce computational resource usage. Extensive experiments validate that our method generalizes to a variety of datasets and tasks and is compatible with a wide range of deep learning architectures and modules. Code is publicly available at https://github.com/lujiazho/ssProp.