Skip to content

tsinghua-fib-lab/SIDSL

Repository files navigation

Official implementation of our CIKM 2025 work "Structure-prior Informed Diffusion Model for Graph Source Localization with Limited Data"

run code

run command:

python main_newdiff.py --dataset [dataset]

Available datasets:(as in datasets\synthesis)

Tunable params can be checked in the following description.

load saved model to run code

run command:

python main_newdiff.py --dataset [dataset] --save_dict [path to saved model]

e.g.

python main_newdiff.py --dataset jazz_IC50 --state_dict ./saved_diffusers/jazz_IC50_best.pt

Requirements

Please check requirements.txt for the required packages.

Main Parameters

  • dataset: Name of the dataset to use
  • gnn_type: Type of GNN architecture (default: gcn)
  • noise_emb_dim: Dimension of noise embedding (default: 128)
  • hidden_dim: Hidden layer dimension (default: 128)
  • num_layers: Number of GNN layers (default: 5)
  • activation: Activation function (default: prelu)
  • mlp_layers: Number of MLP layers (default: 4)
  • num_advisors: Number of advisor models (default: 1)

Optional Parameters

  • feat_drop: Feature dropout rate (default: 0.0)
  • attn_drop: Attention dropout rate (default: 0.0)
  • negative_slope: Negative slope for LeakyReLU (default: 0.2)
  • residual: Whether to use residual connections (default: True)
  • scheduler: Whether to use learning rate scheduler (default: True)
  • train_cond: Whether to train the condition module (default: False)

Cite this paper

@article{chen2025structure,
  title={Structure-prior Informed Diffusion Model for Graph Source Localization with Limited Data},
  author={Chen, Hongyi and Ding, Jingtao and Liang, Xiaojun and Li, Yong and Zhang, Xiao-Ping},
  journal={arXiv e-prints},
  pages={arXiv--2502},
  year={2025}
}

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages