Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
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.
Taxonomy of reduction matrices for Graph Coarsening
Authors: Antonin Joly, Nicolas Keriven, Aline Roumy
NeurIPS 2025 | Venue PDF | LLM Run Details | Input Tokens: 25,113 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 4,822 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare these different choices, both in terms of RSA and performance when used within GNNs trained on coarsened graphs. In this section, we evaluate numerically the performance of these examples, both in terms of RSA constant and used within GNNs. |
| Researcher Affiliation | Academia | Antonin Joly CNRS, IRISA, Rennes, FRANCE EMAIL Nicolas Keriven CNRS, IRISA, Rennes, FRANCE EMAIL Aline Roumy INRIA, Rennes, FRANCE EMAIL |
| Pseudocode | Yes | Algorithm 1 Loukas Algorithm |
| Open Source Code | Yes | The code is available at https://gitlab.inria.fr/anjoly/taxonomy-coarsening-matrices |
| Open Datasets | Yes | We consider the two classical medium-scale graphs Cora [30], and Citeseer [15], and use the public split from [39] for training the GNNs. |
| Dataset Splits | Yes | We consider the two classical medium-scale graphs Cora [30], and Citeseer [15], and use the public split from [39] for training the GNNs. Each training is averaged on 10 random split, using the same experimental setting as in [23], and the hyperparameters are provided in App. G.1. Table 2: Characteristics of Cora and Cite Seer Datasets Dataset # Nodes # Edges # Train Nodes # Val Nodes # Test Nodes Cora 2,708 10,556 140 500 1,000 Cora PCC 2,485 10,138 122 459 915 Citeseer 3,327 9,104 120 500 1,000 Citeseer PCC 2,120 7,358 80 328 663 |
| Hardware Specification | No | handling larger graphs presents challenges. Indeed, in the optimizations we propose, the RSA requires computing the square root of the original Laplacian, which is not sparse in general and, for graphs like Reddit [16], cannot be stored on modern GPUs. |
| Software Dependencies | No | It only uses open-source Python libraries. |
| Experiment Setup | Yes | G.1 Hyperparameters for Tab. 1 For the GCN, for both Cora and Cite Seeer we have 3 convolutional layers with the hidden dimensions [256, 128]. We use an Adam Optimizer with a learning rate lr = 0.01 and a weight decay wd = 0.001. For the SGC model on Cora and Citeseer we make 2 propagations as preprocessing for the features. We use an Adam Optimizer with a learning rate lr = 0.1 and a weight decay wd = 0.001. |