DADAO: Decoupled Accelerated Decentralized Asynchronous Optimization

Authors: Adel Nabli, Edouard Oyallon

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

Reproducibility Variable Result LLM Response
Research Type Experimental Thus, we simultaneously obtain an accelerated rate for both computations and communications, leading to an improvement over stateof-the-art works, our simulations further validating the strength of our relatively unconstrained method. All our experiments are reproducible, using Py Torch (Paszke et al., 2019), our code being online 1.
Researcher Affiliation Academia 1Sorbonne Université, CNRS, ISIR, Paris, France 2Mila, Concordia University, Montréal, Canada.
Pseudocode Yes Algorithm 1: This algorithm block describes our implementation on each local machine. The ODE routine is described by Eq. 8. Algorithm 2: Pseudo-code of our implementation of DADAO on a single machine.
Open Source Code Yes All our experiments are reproducible, using Py Torch (Paszke et al., 2019), our code being online 1. https://github.com/Adel Nabli/DADAO/
Open Datasets No We perform the empirical risk minimization for the decentralized linear regression task given by: Pj=1 aijx cij 2, (9) where aij Rd, and cij R correspond to m local data points stored at node i. We follow a protocol similar to (Kovalev et al., 2021a): we generate n independent synthetic datasets with the make regression functions of scikit-learn (Pedregosa et al., 2011), each worker storing m = 100 data points.
Dataset Splits No The paper states that synthetic datasets were generated and data points per worker, but no explicit train/validation/test splits with percentages or counts are provided. It mentions using 'standard experimental setting' but no specific split information.
Hardware Specification No This work was granted access to the HPC/AI resources of IDRIS under the allocation AD011013743 made by GENCI. (This mentions a resource type but no specific hardware models, CPUs, GPUs, or memory details.)
Software Dependencies No All our experiments are reproducible, using Py Torch (Paszke et al., 2019), our code being online 1. (This cites PyTorch with its original publication year, but does not provide a specific version number like 'PyTorch 1.9' for the software used in the experiments.)
Experiment Setup No We systematically used the proposed hyper-parameters of each reference paper for our implementation without any specific fine-tuning. (This indicates hyperparameters were used, but they are not explicitly listed or described for the current experiments within the paper's text.)