Adapt to Adaptation: Learning Personalization for Cross-Silo Federated Learning

Authors: Jun Luo, Shandong Wu

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluate our method s convergence and generalization behaviors, and perform extensive experiments on two benchmark datasets and two medical imaging datasets under two non-IID settings.
Researcher Affiliation Academia Jun Luo1 , Shandong Wu1,2,3,4 1Intelligent Systems Program, University of Pittsburgh 2Department of Radiology, University of Pittsburgh 3Department of Biomedical Informatics, University of Pittsburgh 4Department of Bioengineering, University of Pittsburgh jul117@pitt.edu, wus3@upmc.edu
Pseudocode Yes Algorithm 1 APPLE
Open Source Code Yes The code is publicly available at https: //github.com/ljaiverson/p FL-APPLE.
Open Datasets Yes Datasets. We use four public datasets including two benchmark datasets: MNIST and CIFAR10, and two medical imaging datasets from the Med MNIST datasets collection [Yang et al., 2021], namely the Organ MNIST(axial) dataset: an 11-class of liver tumor image dataset, and the Path MNIST dataset: a 9-class colorectal cancer image dataset.
Dataset Splits No The paper mentions partitioning datasets into a "training set and a test set" and discusses training for a certain number of "rounds" and "local epochs," but it does not explicitly provide details about a validation set split (e.g., percentages or counts).
Hardware Specification No The paper mentions using "the Bridges-2 system... at the Pittsburgh Supercomputing Center" but does not specify any particular CPU models, GPU models, or detailed hardware configurations (e.g., memory, specific processor types) used for the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We train each method 160 rounds with 5 local epochs and summarize the results as follows. In Equation 7, λ is a dynamic function ranging from 0 and 1, with respect to the round number, r, and µ is a scalar coefficient for the proximal term. More details regarding the loss scheduler are presented in Appendix A.1.