Decoupled Greedy Learning of CNNs for Synchronous and Asynchronous Distributed Learning
This addresses inefficiencies in distributed neural network training for researchers and practitioners, offering a method to parallelize and asynchronize updates, though it is incremental as it builds on existing greedy learning approaches.
The paper tackles the update locking problem in neural network training by proposing Decoupled Greedy Learning (DGL), which decouples layer training for parallelization and extends to asynchronous settings with replay buffers and online vector quantization to reduce communication and memory. It demonstrates effectiveness on CIFAR-10 and ImageNet datasets, showing convergence and competitive performance.
A commonly cited inefficiency of neural network training using back-propagation is the update locking problem: each layer must wait for the signal to propagate through the full network before updating. Several alternatives that can alleviate this issue have been proposed. In this context, we consider a simple alternative based on minimal feedback, which we call Decoupled Greedy Learning (DGL). It is based on a classic greedy relaxation of the joint training objective, recently shown to be effective in the context of Convolutional Neural Networks (CNNs) on large-scale image classification. We consider an optimization of this objective that permits us to decouple the layer training, allowing for layers or modules in networks to be trained with a potentially linear parallelization. With the use of a replay buffer we show that this approach can be extended to asynchronous settings, where modules can operate and continue to update with possibly large communication delays. To address bandwidth and memory issues we propose an approach based on online vector quantization. This allows to drastically reduce the communication bandwidth between modules and required memory for replay buffers. We show theoretically and empirically that this approach converges and compare it to the sequential solvers. We demonstrate the effectiveness of DGL against alternative approaches on the CIFAR-10 dataset and on the large-scale ImageNet dataset.