Bayesian Nonparametric Federated Learning of Neural Networks

Authors: Mikhail Yurochkin, Mayank Agarwal, Soumya Ghosh, Kristjan Greenewald, Nghia Hoang, Yasaman Khazaeni

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We thoroughly evaluate the proposed model and demonstrate its utility empirically in Section 4.
Researcher Affiliation Collaboration 1IBM Research, Cambridge 2MIT-IBM Watson AI Lab 3Center for Computational Health. Correspondence to: Mikhail Yurochkin <mikhail.yurochkin@ibm.com>.
Pseudocode Yes Algorithm 1 Single Layer Neural Matching
Open Source Code Yes Code is available at https://github.com/IBM/ probabilistic-federated-neural-matching
Open Datasets Yes To verify our methodology we simulate federated learning scenarios using two standard datasets: MNIST and CIFAR10.
Dataset Splits No The paper mentions partitioning datasets into 'J batches' for federated learning simulations but does not specify global training/validation/test splits or a separate validation set for model selection, only 'Test Accuracy'.
Hardware Specification No The paper does not provide specific details on the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper does not specify version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes In this and subsequent experiments each local neural network has Lj = 100 hidden neurons. We compare PFNM, using the communication procedure from Section 3.3 (σ = σ0 = γ0 = 1 across experiments) to federated averaging and the distributed optimization approach, downpour SGD (D-SGD) of Dean et al. (2012). Further, k-Means requires us to choose k, which we set to K = min(500, 50J).