EControl: Fast Distributed Optimization with Compression and Error Control

Authors: Yuan Gao, Rustem Islamov, Sebastian U Stich

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive numerical evaluations to illustrate the efficacy of our method and support our theoretical findings.
Researcher Affiliation Collaboration 1Universität des Saarlandes, 2Universität Basel, 3CISPA Helmholtz Center for Information Security
Pseudocode Yes Algorithm 1 EC-Ideal; Algorithm 2 EControl
Open Source Code Yes We provide source code as part of the supplementary material which allows to reproduce Deep Learning and synthetic least squares experiments.
Open Datasets Yes MNIST dataset (Deng, 2012); Cifar-10 (Krizhevsky et al., 2014) dataset.
Dataset Splits No The paper mentions 'train (90%) and test (10%) sets' for the Logistic Regression problem and for Cifar-10, but no explicit validation set split is described.
Hardware Specification No No specific hardware (e.g., GPU/CPU models, memory details, or cloud instance types) used for experiments is mentioned in the paper.
Software Dependencies No The implementation is done in Pytorch (Paszke et al., 2019) but no specific version number for PyTorch or other software dependencies is provided.
Experiment Setup Yes Stepsizes were tuned for each setting. X-axis represents the number of bits sent. ... For EControl we fine-tune η over {10^−3, 5 × 10^−3, 10^−2, 5 × 10^−2, 10^−1}. ... We fine-tune the stepsizes of the methods over {1, 10^−1, 10^−2, 10^−3}. Moreover, we fine-tune η parameter for EControl over {0.2, 0.1, 0.05}.