Asynchronous Stochastic Gradient Descent with Delay Compensation
This addresses the efficiency-stability trade-off in distributed deep learning training, representing an incremental improvement to existing ASGD methods.
The paper tackles the problem of delayed gradients in asynchronous stochastic gradient descent (ASGD) by proposing a delay compensation technique using Taylor expansion and Hessian approximation, resulting in DC-ASGD which outperforms synchronous and asynchronous SGD and nearly matches sequential SGD performance on CIFAR-10 and ImageNet datasets.
With the fast development of deep learning, it has become common to learn big neural networks using massive training data. Asynchronous Stochastic Gradient Descent (ASGD) is widely adopted to fulfill this task for its efficiency, which is, however, known to suffer from the problem of delayed gradients. That is, when a local worker adds its gradient to the global model, the global model may have been updated by other workers and this gradient becomes "delayed". We propose a novel technology to compensate this delay, so as to make the optimization behavior of ASGD closer to that of sequential SGD. This is achieved by leveraging Taylor expansion of the gradient function and efficient approximation to the Hessian matrix of the loss function. We call the new algorithm Delay Compensated ASGD (DC-ASGD). We evaluated the proposed algorithm on CIFAR-10 and ImageNet datasets, and the experimental results demonstrate that DC-ASGD outperforms both synchronous SGD and asynchronous SGD, and nearly approaches the performance of sequential SGD.