Federated Learning with Manifold Regularization and Normalized Update Reaggregation

Authors: Xuming An, Li Shen, Han Hu, Yong Luo

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate that Fed MRUR can achieve a new state-of-the-art (SOTA) accuracy with less communication. In this section, we validate the effectiveness of the proposed Fed MRUR algorithm using the experimental results on CIFAR-10/100 [18] and Tiny Image Net [18].
Researcher Affiliation Collaboration 1 School of Information and Electronics, Beijing Institute of Technology, China 2 JD Explore Academy, China 3 School of Computer Science, Wuhan University, China
Pseudocode Yes Algorithm 1 Fed MRUR
Open Source Code No The paper does not contain an explicit statement about releasing open-source code or a link to a repository for the described methodology.
Open Datasets Yes In this section, we validate the effectiveness of the proposed Fed MRUR algorithm using the experimental results on CIFAR-10/100 [18] and Tiny Image Net [18].
Dataset Splits Yes The CIFAR-10 dataset consists of 50K training images and 10K testing images. All the images are with 32x32 resolution belonging to 10 categories. In the CIFAR-100 dataset, there are 100 categories of images with the same format as CIFAR-10. Tiny Image Net includes 200 categories of 100K training images and 10K testing images, whose resolutions are 64x64.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions software components like 'Res Net-18' and 'SAM optimizer' but does not provide specific version numbers for any software dependencies (e.g., PyTorch, TensorFlow, Python, CUDA versions).
Experiment Setup Yes For all algorithms on all datasets, following [1, 42], the local and global learning rates are set as 0.1 and 1.0, the learning rate decay is set as 0.998 per communication round and the weight decay is set as 5e-4. Res Net-18 together with group normalization is adopted as the backbone to train the model. The clients settings for different tasks are summarized in Table 1. Other optimizer hyperparameters are as follow: ρ = 0.5 for SAM, α = 0.1 for client momentum, γ = 0.005, σ = 10000.0 and β = 1 for manifold regularization.