One Forward is Enough for Neural Network Training via Likelihood Ratio Method

TL;DR

This paper introduces a likelihood ratio method for neural network training that requires only one forward pass, offering greater flexibility and efficiency compared to traditional backpropagation.

Contribution

The authors propose a unified likelihood ratio approach for gradient estimation, eliminating recursive backpropagation and enabling flexible architecture design and device adaptation.

Findings

ULR achieves effective training with a single forward pass.

The method improves training flexibility and robustness.

Variance reduction techniques accelerate convergence.

Abstract

While backpropagation (BP) is the mainstream approach for gradient computation in neural network training, its heavy reliance on the chain rule of differentiation constrains the designing flexibility of network architecture and training pipelines. We avoid the recursive computation in BP and develop a unified likelihood ratio (ULR) method for gradient estimation with just one forward propagation. Not only can ULR be extended to train a wide variety of neural network architectures, but the computation flow in BP can also be rearranged by ULR for better device adaptation. Moreover, we propose several variance reduction techniques to further accelerate the training process. Our experiments offer numerical results across diverse aspects, including various neural network training scenarios, computation flow rearrangement, and fine-tuning of pre-trained models. All findings demonstrate that ULR effectively enhances the flexibility of neural network training by permitting localized module training without compromising the global objective and significantly boosts the network robustness.