Personalized Federated Learning With Gaussian Processes

Authors: Idan Achituve, Aviv Shamsian, Aviv Navon, Gal Chechik, Ethan Fetaya

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on standard PFL benchmarks with CIFAR-10, CIFAR-100, and CINIC-10, and on a new setup of learning under input noise show that p Fed GP achieves well-calibrated predictions while significantly outperforming baseline methods, reaching up to 21% in accuracy gain. 6 Experiments We evaluated p Fed GP against baseline methods in various learning setups.
Researcher Affiliation Collaboration Idan Achituve Bar-Ilan University, Israel idan.achituve@biu.ac.il Aviv Shamsian Bar-Ilan University, Israel aviv.shamsian@biu.ac.il Aviv Navon Bar-Ilan University, Israel aviv.navon@biu.ac.il Gal Chechik Bar-Ilan University, Israel NVIDIA, Isreal gal.chechik@biu.ac.il Ethan Fetaya Bar-Ilan University, Israel ethan.fetaya@biu.ac.il
Pseudocode Yes Algorithm 1 p Fed GP. C clients indexed by c; E number of local epochs; |S| number of sampled clients; M number of inducing inputs per class
Open Source Code Yes Our code is publicly available at https://github.com/Idan Achituve/p Fed GP
Open Datasets Yes All methods were evaluated on CIFAR-10, CIFAR-100 [38], and CINIC-10 [15] datasets.
Dataset Splits Yes We tuned the hyperparameters of all methods using a pre-allocated held-out validation set.
Hardware Specification No The paper does not provide specific details on the hardware (e.g., GPU models, CPU types, memory) used for running the experiments.
Software Dependencies No The paper does not specify versions for software dependencies or libraries (e.g., Python version, PyTorch/TensorFlow version, specific library versions).
Experiment Setup Yes We limit the training process to 1000 communication rounds, in each we sample five clients uniformly at random for model updates. In the local model, we performed 100 epochs of training for each client. In all experiments, we used a Le Net-based network [40] having two convolution layers followed by two fully connected layers and an additional linear layer.