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 [1].
TopInG: Topologically Interpretable Graph Learning via Persistent Rationale Filtration
Authors: Cheng Xin, Fan Xu, Xin Ding, Jie Gao, Jiaxin Ding
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate TOPING s effectiveness in tackling key challenges, such as handling variform rationale subgraphs, balancing predictive performance with interpretability, and mitigating spurious correlations. Results show that our approach improves upon state-of-the-art methods on both predictive accuracy and interpretation quality. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Rutgers University, Piscataway, NJ, USA 2School of Computer Science, Shanghai Jiao Tong University, Shanghai, China. Correspondence to: Cheng Xin <EMAIL>, Jiaxin Ding <EMAIL>. |
| Pseudocode | No | The paper describes methods using prose and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing the source code for the methodology described, nor does it provide a link to a code repository. |
| Open Datasets | Yes | Datasets. We consider eight benchmark datasets commonly used in the graph explainability literature, categorized into three types: Single Motif, Multiple Motif, and Real Dataset. The first two consist of synthetic datasets. Single Motif includes BA-2Motifs (Luo et al., 2020), BAHouse Grid (Amara et al., 2023), SPmotif0.5 and SPmotif0.9 (Wu et al., 2022). These datasets contain graphs with a single type of motif or structural pattern repeated throughout. Multiple Motif includes BA-House And Grid, BA-House Or Grid (Bui et al., 2024), and BA-House Or Gridn Rnd. The last one is a synthetic dataset we create for verifying the variform rationale challenge for existing intrinsic methods (see Appendix E for more details). Real Dataset include Mutag (Luo et al., 2020) and Benzene (Sanchez-Lengeling et al., 2020). |
| Dataset Splits | Yes | Data Splits. For BA synthetic datasets, we follow the previous work (Miao et al., 2022; Chen et al., 2024; Bui et al., 2024) to split them into three sets(80%/10%/10%). For SPmotifs and real datasets, we use the default splits. |
| Hardware Specification | Yes | All experiments were conducted on a single RTX 4090 GPU. |
| Software Dependencies | No | The paper mentions using codebase from (Zhang et al., 2022) for computations, but does not provide specific version numbers for software dependencies such as Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | Backbone Architecture. We use a two-layer GIN (Xu et al., 2019) with 64 hidden dimensions and 0.3 dropout ratio for all baselines. We use a three-layer CINpp (Giusti et al., 2023) with 64 hidden dimensions and 0.15/0.3 dropout ratio for TOPING. ... All the results are averaged over 5 runs tested with different random seeds. ... The ฮป of information regularizer is set to be 1. As for topological constraint, we set the coefficient to 0.01 to achieve the best performance, which aligning with Figure 4. In practice, we simply fix w = 0.5, ยต1 = 0.25, ยต2 = 0.75, and initialize r1 = r2 = 0.25. |