Efficient Deep Learning with Decorrelated Backpropagation

2024-05-03

arXiv.org

AI Summary

To address slow convergence, high computational overhead, and the difficulty of balancing accuracy and efficiency caused by input correlations in large-scale deep neural network (DNN) training, this paper proposes Decorrelated Backpropagation (DeCorrBP)—a lightweight, end-to-end input decorrelation method that requires no architectural modifications or additional parameters. Its core components include a gradient covariance suppression mechanism, a computationally efficient inter-layer whitening approximation, and seamless integration into standard training pipelines. Experiments on ResNets up to 50 layers demonstrate over 2× training speedup, improved test accuracy, and substantial reductions in GPU-hours and carbon footprint. To our knowledge, DeCorrBP is the first practical, stable, and efficient end-to-end input decorrelation method applicable to large-scale DNNs.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Learning & Optimization for CV
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The backpropagation algorithm remains the dominant and most successful method for training deep neural networks (DNNs). At the same time, training DNNs at scale comes at a significant computational cost and therefore a high carbon footprint. Converging evidence suggests that input decorrelation may speed up deep learning. However, to date, this has not yet translated into substantial improvements in training efficiency in large-scale DNNs. This is mainly caused by the challenge of enforcing fast and stable network-wide decorrelation. Here, we show for the first time that much more efficient training of deep convolutional neural networks is feasible by embracing decorrelated backpropagation as a mechanism for learning. To achieve this goal we made use of a novel algorithm which induces network-wide input decorrelation using minimal computational overhead. By combining this algorithm with careful optimizations, we achieve a more than two-fold speed-up and higher test accuracy compared to backpropagation when training several deep networks up to a 50-layer ResNet model. This demonstrates that decorrelation provides exciting prospects for efficient deep learning at scale.

Problem

• Speeding up deep learning via input decorrelation
• Reducing computational cost in large-scale DNN training
• Enhancing training efficiency and accuracy in deep networks

Innovation

• Decorrelated backpropagation enhances deep learning efficiency
• Novel algorithm enforces network-wide input decorrelation
• Combined optimizations yield faster training and accuracy