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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Graph Persistence goes Spectral
Authors: Mattie Ji, Amauri H. Souza, Vikas Garg
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, experiments on synthetic and real-world datasets demonstrate the effectiveness of Spect Re and its potential to enhance the capabilities of graph models in relevant learning tasks. Code is available at https://github.com/Aalto-Qu ML/Spect Re/. 1 Introduction ... To validate our theoretical analysis and show the effectiveness of Spect Re diagrams, we conduct experiments using two sets of synthetic datasets for assessing the expressive power of graph models (13 datasets in total), and multiple real datasets. |
| Researcher Affiliation | Collaboration | Mattie Ji University of Pennsylvania EMAIL Amauri H. Souza Federal Institute of Ceará EMAIL Vikas Garg Aalto University Yai Yai Ltd EMAIL |
| Pseudocode | No | The paper only refers to algorithms in other papers (e.g., Algorithm 1 and 2 of [33]) but does not provide any pseudocode or algorithm blocks within its own content. |
| Open Source Code | Yes | Code is available at https://github.com/Aalto-Qu ML/Spect Re/. |
| Open Datasets | Yes | To assess the effectiveness of Spect Re from an empirical perspective, we consider two sets of experiments. The first consists of isomorphism tests on synthetic benchmarks designed to evaluate the expressive power of graph models. The second explores the combination of topological descriptors and GNNs in real-world tasks. Implementation details are provided in Appendix E. ... MUTAG, PTC-MM, PTC-MR, PTC-FR, NCI1, NCI109, IMDB-B, ZINC, MOLHIV [42]. ... The datasets are available at https://chrsmrrs.github.io/datasets/ docs/datasets/. In addition, MOLHIV is the largest dataset (over 41K graphs) and is part of the Open Graph Benchmark1. We also consider a regression task using the ZINC dataset a subset of the popular ZINC-250K chemical compounds [34] |
| Dataset Splits | Yes | We use a random 80%/10%/10% (train/val/test) split for all datasets. All models are initialized with a learning rate of 10 3 that is halved if the validation loss does not improve over 10 epochs. We apply early stopping with patience equal to 30. For ZINC and OGB-MOLHIV, we use public splits. |
| Hardware Specification | Yes | For all experiments, we use a cluster with Nvidia V100 GPUs. |
| Software Dependencies | No | We implement all models using the Py Torch Geometric Library [23]. |
| Experiment Setup | Yes | We set the number of hidden units in the Deep Set and GNN layers to 32, and of the filtration functions to 16 i.e., the vertex/edge filtration functions consist of a 2-layer MLP with 16 hidden units. The GNN node embeddings are combined using a global mean pooling layer. We employ the Adam optimizer [35] with a maximum of 500 epochs, learning rate of 10 4, and batch size equal to 64. We use a random 80%/10%/10% (train/val/test) split for all datasets. All models are initialized with a learning rate of 10 3 that is halved if the validation loss does not improve over 10 epochs. We apply early stopping with patience equal to 30. |