State Space Models on Temporal Graphs: A First-Principles Study

Authors: Jintang Li, Ruofan Wu, Xinzhou Jin, Boqun Ma, Liang Chen, Zibin Zheng

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this work, we undertake a principled investigation that extends SSM theory to temporal graphs by integrating structural information into the online approximation objective via the adoption of a Laplacian regularization term. The emergent continuous-time system introduces novel algorithmic challenges, thereby necessitating our development of GRAPHSSM, a graph state space model for modeling the dynamics of temporal graphs. Extensive experimental results demonstrate the effectiveness of our GRAPHSSM framework across various temporal graph benchmarks.
Researcher Affiliation Collaboration Jintang Li1 , Ruofan Wu2 , Xinzhou Jin1, Boqun Ma3, Liang Chen1 , Zibin Zheng1 1Sun Yat-sen University, 2Coupang, 3Shanghai Jiao Tong University
Pseudocode Yes A detailed exposition of the GRAPHSSM-S4 (resp. GRAPHSSM-S5, GRAPHSSM-S6) layer is provided in algorithm 1 (resp. algorithm 2, algorithm 3) in appendix D.2.
Open Source Code Yes Code will be made available at https: //github.com/Edison Leeeee/Graph SSM.
Open Datasets Yes The experiments are conducted on six temporal graph benchmarks with different scales and time snapshots, including DBLP-3 [47], Brain [47], Reddit [47], DBLP-10 [25], ar Xiv [19], and Tmall [25].
Dataset Splits Yes For the DBLP-3, Brain, and Reddit datasets, we adopt the 81%/9%/10% train/validation/test splits as suggested in [47].
Hardware Specification Yes All experiments are conducted on an NVIDIA RTX 3090 Ti GPU with 24 GB memory.
Software Dependencies No The paper states: 'We implement our models as well as baselines with Py Torch [35] and Py Torch Geometric [5]'. However, it does not specify version numbers for these software dependencies, which is required for reproducibility.
Experiment Setup Yes For the DBLP-3, Brain, and Reddit datasets, we adopt the 81%/9%/10% train/validation/test splits as suggested in [47]. For the DBLP-10 and Tmall datasets, we follow the experimental settings of previous works [25], where 80% of the nodes are randomly selected as the training set, and the remaining nodes are used as the test set. We employ feature mixing for DBLP-10 and representation mixing for other datasets. The graph convolution networks used to learn the graph structure are SAGE [16] for all datasets, except for ar Xiv, where GCN [21] is used.