Gradient Sparsification for Communication-Efficient Distributed Optimization
This addresses communication overhead in distributed machine learning, but it is incremental as it builds on existing sparsification techniques.
The paper tackles the communication bottleneck in distributed optimization by proposing gradient sparsification methods to reduce coding length, with experiments validating effectiveness on tasks like logistic regression and CNNs.
Modern large scale machine learning applications require stochastic optimization algorithms to be implemented on distributed computational architectures. A key bottleneck is the communication overhead for exchanging information such as stochastic gradients among different workers. In this paper, to reduce the communication cost we propose a convex optimization formulation to minimize the coding length of stochastic gradients. To solve the optimal sparsification efficiently, several simple and fast algorithms are proposed for approximate solution, with theoretical guaranteed for sparseness. Experiments on $\ell_2$ regularized logistic regression, support vector machines, and convolutional neural networks validate our sparsification approaches.