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].
Compositional PAC-Bayes: Generalization of GNNs with persistence and beyond
Authors: Kirill Brilliantov, Amauri Souza, Vikas Garg
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical evaluations on several standard real-world datasets demonstrate that our theoretical bounds highly correlate with empirical generalization performance, leading to improved classifier design via our regularizers. Overall, this work bridges a crucial gap in the theoretical understanding of PH methods and general heterogeneous models, paving the way for the design of better models for (graph) representation learning. Our code is available at https://github.com/Aalto-Qu ML/Compositional-PAC-Bayes. |
| Researcher Affiliation | Collaboration | Kirill Brilliantov ETH Zürich EMAIL Amauri H. Souza Federal Institute of Ceará EMAIL Vikas Garg Yai Yai Ltd & Aalto University EMAIL |
| Pseudocode | No | No sections or figures explicitly labeled 'Pseudocode' or 'Algorithm' were found. |
| Open Source Code | Yes | Our code is available at https://github.com/Aalto-Qu ML/Compositional-PAC-Bayes. |
| Open Datasets | Yes | We use six popular benchmarks for graph classification: DHFR, MUTAG, PROTEINS, NCI1, IMDB-BINARY, MOLHIV, which are available as part of TUDatasets [26] and OGB [21]. |
| Dataset Splits | Yes | We use a 80/10/10% (train/val/test) split for all datasets when we perform model selection. |
| Hardware Specification | Yes | Hardware. For all experiments, we use Tesla V100 GPU cards and consider a memory budget of 32GB of RAM. |
| Software Dependencies | No | The paper mentions 'Py Torch [42]' but does not specify its version number or other software dependencies with versions. |
| Experiment Setup | Yes | All models are trained with Adam [27] and learning rate of 10 3 for 3000 epochs. For the experiments with GNNs, we kept only the larger datasets (and added results for the NCI109 dataset). Regarding filtration functions, we closely follow [5] and use Heat kernels with parameter values equal to 0.1 and 10. |