Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Subgraph Federated Learning via Spectral Methods
Authors: Javad Aliakbari, Johan Oestman, Ashkan Panahi, Alexandre Graell i Amat
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on benchmark datasets demonstrate that FEDLAP achieves competitive or superior utility compared to existing techniques. We provide a formal analysis of the privacy of FEDLAP, demonstrating that it preserves privacy. Notably, FEDLAP is the first subgraph FL scheme with strong privacy guarantees. Extensive experiments on benchmark datasets demonstrate that FEDLAP achieves competitive or superior utility compared to existing techniques. |
| Researcher Affiliation | Collaboration | Javad Aliakbari1 Johan Östman2 Ashkan Panahi1 Alexandre Graell i Amat1 1Chalmers University of Technology 2AI Sweden |
| Pseudocode | Yes | Algorithm 1 The Arnoldi iteration for the computation of an orthonormal basis of a Krylov space Algorithm 2 The Decentralized Arnoldi algorithm for the computation of an orthonormal basis of a Krylov space |
| Open Source Code | Yes | The code is available at this link. |
| Open Datasets | Yes | Experiments are conducted on six datasets: Cora and Citeseer [26], Pub Med [27], Chameleon [28], Amazon Photo [29], and Ogbn-Arxiv [30]. |
| Dataset Splits | Yes | Table 2: Node classification accuracy with random partitioning. Nodes are split into train-val-test as 10%-10%-80%. For each result, the mean and standard deviation are shown for 10 independent runs. |
| Hardware Specification | Yes | Our experiments were conducted on a machine with 2 NVIDIA Tesla V100 SXM2 GPUs, each with 32GB of RAM. |
| Software Dependencies | No | The paper discusses various methods and libraries but does not provide specific version numbers for the software environment used for implementation (e.g., Python, PyTorch, CUDA versions are not mentioned). |
| Experiment Setup | Yes | In the following we provide the hyperparameters used in the experiments, obtained through a grid search to optimize performance. In particular, Table 4 contains, for the different datasets, the learning rate λ, the weight decay in the L2 regularization, the number of training iterations (epochs), the regularization parameter λreg, the dimensionality of the NSFs, ds, the truncation number r, and the model architecture of the node feature and node structure feature predictors, f θf and gθs, respectively. |