Momentum Ensures Convergence of SIGNSGD under Weaker Assumptions

Authors: Tao Sun, Qingsong Wang, Dongsheng Li, Bao Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on training Res Net110 from (He et al., 2016) for CIFAR-100 (Krizhevsky et al., 2009) classification using SIGNSGD and SIGNSGD-SIM with various batch sizes.
Researcher Affiliation Academia 1College of Computer, National University of Defense Technology, Hunan, China. 2University of Utah.
Pseudocode Yes Algorithm 1 SIGNSGD with SImple Momentum (SIGNSGD-SIM)
Open Source Code No Our code is based on open-source libraries2. 2github.com/akamaster/pytorch_resnet_ cifar10, github.com/epfml/error-feedback-SGD - The text states their code is based on open-source libraries and provides links to those libraries, but does not explicitly state their own implementation code is open-source or provide a link to it.
Open Datasets Yes We train various Res Net models from (He et al., 2016) on CIFAR-10/CIFAR-100 (Krizhevsky et al., 2009)
Dataset Splits No Both datasets are split into a training set of 50,000 images and a test set of 10,000 images. The paper mentions training and test sets, and data distribution for clients, but does not explicitly describe a validation set or its split.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running experiments were provided.
Software Dependencies No Our code is based on open-source libraries2. 2github.com/akamaster/pytorch_resnet_ cifar10, github.com/epfml/error-feedback-SGD - The paper mentions libraries but does not provide specific version numbers for them.
Experiment Setup Yes The learning rate is decimated twice during this time, first at 100 epochs and again at 150 epochs. The initial learning rate for a batch size of 128 is 1 10 3. (...) The momentum parameter of SIGNSGD-SIM is set to 0.9, and the weight decay for both algorithms is set to 1 10 4.