FedSSP: Federated Graph Learning with Spectral Knowledge and Personalized Preference

Authors: Zihan Tan, Guancheng Wan, Wenke Huang, Mang Ye

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform extensive experiments on cross-dataset and cross-domain settings to demonstrate the superiority of our framework. Furthermore, We perform extensive experiments on cross-dataset and cross-domain settings to demonstrate the superiority of our framework.
Researcher Affiliation Academia 1 National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, Hubei Key Laboratory of Multimedia and Network Communication Engineering, School of Computer Science, Wuhan University, Wuhan, China. 2 Taikang Center for Life and Medical Sciences, Wuhan University, Wuhan, China
Pseudocode No The paper provides architectural diagrams (Figure 2) and mathematical formulations, but no explicit pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https://github.com/Oakley Tan/Fed SSP.
Open Datasets Yes Follow the settings in [61], we use 15 public graph classification datasets from four different domains, including Small Molecules (MUTAG, BZR, COX2, DHFR, PTC_MR, AIDS, NCI1), Bioinformatics (PROTEIN, OHSU, Peking_1), Social Networks(IMDB-BINARY, IMDBMULTI), and Computer Vision (Letter-low, Letter-high, Letter-med) [52].
Dataset Splits Yes For each setting, every client holds its unique graph dataset, among which 10% are held out for testing, 10% for validation, and 80% for training.
Hardware Specification Yes The experiments are conducted using NVIDIA Ge Force RTX 3090 GPUs as the hardware platform, coupled with Intel(R) Xeon(R) Gold 6240 CPU @ 2.60GHz.
Software Dependencies No The paper mentions leveraging the Adam W optimizer but does not specify specific version numbers for software libraries (e.g., PyTorch, TensorFlow) or other key software components used for the experiments.
Experiment Setup Yes We leverage the Adam W optimizer [31] for local GNNs with learning rate 0.001, the default parameter of ϵ = 1e 8, and (β1, β2) = (0.99, 0.999), as suggested by [54, 85]. The number of communication rounds is 200 for all FL methods. For results in Tab. 1, we set up 4 heads for the multi-head attention mechanism while 128 for the hidden dimension.