Gradient Sparsification for Communication-Efficient Distributed Optimization

Authors: Jianqiao Wangni, Jialei Wang, Ji Liu, Tong Zhang

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we conduct experiments to validate the effectiveness and efficiency of the proposed sparsification technique.
Researcher Affiliation Collaboration Jianqiao Wangni University of Pennsylvania Tencent AI Lab wnjq@seas.upenn.edu Jialei Wang Two Sigma Investments jialei.wang@twosigma.com Ji Liu University of Rochester Tencent AI Lab ji.liu.uwisc@gmail.com Tong Zhang Tencent AI Lab tongzhang@tongzhang-ml.org
Pseudocode Yes Algorithm 1 A synchronous distributed optimization algorithm, Algorithm 2 Closed-form solution, Algorithm 3 Greedy algorithm
Open Source Code No The paper does not provide concrete access to source code for the methodology described. The provided link (https://arxiv.org/abs/1710.09854) is to a full version of the paper, not a code repository.
Open Datasets Yes We consider the convolutional neural networks (CNN) on the CIFAR-10 dataset with different settings.
Dataset Splits No The paper mentions using CIFAR-10, which has standard splits, and synthetic data, but does not explicitly provide specific percentages, sample counts, or citations to predefined splits for training, validation, or testing.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments, only mentioning a 'shared memory multi-thread' setup.
Software Dependencies No The paper mentions specific optimization algorithms like ADAM and SGD, but does not provide specific software dependencies with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x, Python 3.x).
Experiment Setup Yes The mini-batch size is set to be 8 by default unless otherwise specified. and The step sizes are fine-tuned on each case... and The initial step size is set to 0.02. and The number of workers is set to 16 or 32, the regularization parameter is set to {0.5, 0.1, 0.05}, and the learning rate is chosen from {0.5, 0.25, 0.05, 0.25}.