Modeling Inter-Intra Heterogeneity for Graph Federated Learning

Official Code Repository for our paper (AAAI 2025) :

Modeling Inter-Intra Heterogeneity for Graph Federated Learning

Highlights

• User-Friendly: Simple and intuitive to use.
• SOTA Performance: Outperforms the second-best by 5.79% on all heterophilic datasets.
• Rich Datasets: Covers 11 GFL datasets (6 homophilic, 5 heterophilic).

Requirement

• Python 3.8.8
• PyTorch 1.12.0+cu113
• PyTorch Geometric 2.3.0
• METIS (only for subgraph generation) https://github.com/james77777778/metis_python

Subgraph generation

Download from the Google Drive (https://drive.google.com/file/d/1PyqvR6yL43Om42fdsbKHj5WCgREvi3St/view?usp=sharing) and then unzip it.

Place the datasets folder in the same path as README.md

or

Follow command lines automatically to generate the subgraphs.

$ cd FedIIH_2/data/generators
$ python disjoint.py
$ python overlapping.py

Parameter description

• gpus: specify gpus to use
• num workers: specify the number of workers on gpus (e.g., if your experiment uses 10 clients for every round then use less than or equal to 10 workers). The actual number of workers will be num_workers + 1 (one additional worker for a server).
• FedIIH_2 means that the number of disentangled latent factors is set to 2 (K=2). Similarly, FedIIH_10 means that the number of disentangled latent factors is set to 10 (K=10).

Homophilic datasets

Follow command lines to run the experiments.

Cora

non-overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Cora --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Cora --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

CiteSeer

non-overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset CiteSeer --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset CiteSeer --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

PubMed

non-overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset PubMed --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset PubMed --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Amazon-Computer

non-overlapping

$ cd FedIIH_6
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Computers --mode disjoint --frac 1.0 --n-rnds 200 --n-eps 3 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Computers --mode overlapping --frac 1.0 --n-rnds 200 --n-eps 3 --n-clients 10 --seed 42

Amazon-Photo

non-overlapping

$ cd FedIIH_6
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Photo --mode disjoint --frac 1.0 --n-rnds 200 --n-eps 2 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_10
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Photo --mode overlapping --frac 1.0 --n-rnds 200 --n-eps 2 --n-clients 10 --seed 42

ogbn-arxiv

non-overlapping

$ cd FedIIH_6
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset ogbn-arxiv --mode disjoint --frac 1.0 --n-rnds 200 --n-eps 2 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_6
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset ogbn-arxiv --mode overlapping --frac 1.0 --n-rnds 200 --n-eps 2 --n-clients 10 --seed 42

Heterophilic datasets

Roman-empire

non-overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Roman-empire --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Roman-empire --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Amazon-ratings

non-overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Amazon-ratings --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Amazon-ratings --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Minesweeper

non-overlapping

$ cd FedIIH_6
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Minesweeper --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_4
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Minesweeper --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Tolokers

non-overlapping

$ cd FedIIH_10
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Tolokers --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_10
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Tolokers --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Questions

non-overlapping

$ cd FedIIH_8
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Questions --mode disjoint --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

overlapping

$ cd FedIIH_2
$ python main.py --gpu 0 --n-workers 1 --model fedhvae --dataset Questions --mode overlapping --frac 1.0 --n-rnds 100 --n-eps 1 --n-clients 10 --seed 42

Citation

If you found our code or our paper useful in your work, please cite our work. Thank you very much!

@inproceedings{yu2025modeling,
  title={Modeling Inter-Intra Heterogeneity for Graph Federated Learning},
  author={Yu, Wentao and Chen, Shuo and Tong, Yongxin and Gu, Tianlong and Gong, Chen},
  booktitle={AAAI Conference on Artificial Intelligence},
  pages={1--7},
  year={2025}
}

Reference

The code structure is inspired by the code in FED-PUB (https://github.com/JinheonBaek/FED-PUB).