Robust Collaborative Learning with Linear Gradient Overhead

Authors: Sadegh Farhadkhani, Rachid Guerraoui, Nirupam Gupta, Lê-Nguyên Hoang, Rafael Pinot, John Stephan

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

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our theory by experiments on image classification and make our code available at https://github.com/LPD-EPFL/ robust-collaborative-learning.
Researcher Affiliation Collaboration 1EPFL, Lausanne, Switzerland. 2Tournesol, 3Calicarpa, Switzerland. Correspondence to: Sadegh Farhadkhani <sadegh.farhadkhani@epfl.ch>.
Pseudocode Yes Algorithm 1 MONNA as executed by a correct node i
Open Source Code Yes We validate our theory by experiments on image classification and make our code available at https://github.com/LPD-EPFL/ robust-collaborative-learning.
Open Datasets Yes We use the MNIST (Le Cun & Cortes, 2010) and CIFAR-10 (Krizhevsky et al., 2009) datasets, pre-processed as in (Baruch et al., 2019) and (El Mhamdi et al., 2021b).
Dataset Splits No The paper mentions using MNIST and CIFAR-10 datasets and discusses data heterogeneity, but it does not explicitly provide details about training, validation, and test splits (e.g., percentages or counts).
Hardware Specification Yes We list below the hardware components used: 1 Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz, 2 Nvidia Ge Force GTX 1080 Ti, 64 GB of RAM
Software Dependencies Yes Python 3.8.10 has been used to run our scripts. Besides the standard libraries associated with Python 3.8.10, our scripts use the following libraries: numpy 1.19.1, torch 1.6.0, torchvision 0.7.0, pandas 1.1.0, matplotlib 3.0.2, requests 2.21.0, urllib3 1.24.1, chardet 3.0.4, certifi 2018.08.24, pytz 2020.1, dateutil 2.6.1, pyparsing 2.2.0, cycler 0.10.0, kiwisolver 1.0.1, cffi 1.13.2.
Experiment Setup Yes For MNIST, we consider a convolutional neural network (CNN)... The model is trained using a learning rate γ = 0.75 for T = 600 iterations. ... For CIFAR-10, we use a CNN... we set γ = 0.5 and T = 2000 iterations. ... Momentum β = 0.99 (except for SCC: β = 0.9), Batch size b = 25 (MNIST), b = 50 (CIFAR-10).