DMbaGCN

This is the official implementation of the following paper:

Dual Mamba for Node-Specific Representation Learning: Tackling Over-Smoothing with Selective State Space Modeling

Accepted by AAAI 2026

Environment Setup

Required Dependencies :

• torch>=2.1.2
• torch_geometric>=2.5.2
• python>=3.8
• einops>=0.8.0
• scipy>=1.12.0
• numpy>=1.23.5

Quick Start

CoraFull Dataset

The main experiments:

cd NodeClassification/

python training.py --dataset Corafull --d_model 512 --d_inner 512 --dt_rank 64 --d_state 1 --mamba_dropout 0.2 --alpha 0.9 --graph_weight 0.8 --layer_num 3 --lr 0.005 --weight_decay 0. --net GCN_mamba_Net --runs 10

Note: The dataset will be automatically downloaded when the code is executed

Citation

If you find our repository useful for your research, please consider citing our paper:

@article{he2025dual,
  title={Dual Mamba for Node-Specific Representation Learning: Tackling Over-Smoothing with Selective State Space Modeling},
  author={He, Xin and Wang, Yili and Dai, Yiwei and Wang, Xin},
  journal={arXiv preprint arXiv:2511.06756},
  year={2025}
}