Unbiased Compression Saves Communication in Distributed Optimization: When and How Much?

Authors: Yutong He, Xinmeng Huang, Kun Yuan

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We support our theoretical findings with experiments on both synthetic data and real datasets.
Researcher Affiliation Academia Yutong He Peking University yutonghe@pku.edu.cn Xinmeng Huang University of Pennsylvania xinmengh@sas.upenn.edu Kun Yuan Peking University kunyuan@pku.edu.cn
Pseudocode Yes Algorithm 1: ADIANA
Open Source Code No The paper does not provide an explicit statement or link to its own open-source code for the methodology described.
Open Datasets Yes Logistic regression. Consider a distributed logistic regression problem (1) with fi(x) := 1 M PM m=1 ln(1 + exp( bi,ma i,mx), where {(ai,m, bi,m)}1 i n,1 m M are datapoints in a9a and w8a datasets from LIBSVM [11].
Dataset Splits No The paper mentions distributing data to nodes for distributed optimization but does not explicitly describe train/validation/test splits for the datasets used in the experiments.
Hardware Specification Yes Computational resource. All experiments are run on an NVIDIA A100 server.
Software Dependencies No The paper mentions implementing algorithms but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes F.3 Parameter values. In this subsection, we list all the parameter values that are selected by applying Bayesian Optimization. Table 2, 3, 4, 5, 6 list the parameters chosen in the least squares problem, logistic regression using a9a dataset, logistic regression using w8a dataset, the constructed problem, and logistic regression using CIFAR-10 dataset, respectively.