Graph Positional Encoding via Random Feature Propagation
Authors: Moshe Eliasof, Fabrizio Frasca, Beatrice Bevilacqua, Eran Treister, Gal Chechik, Haggai Maron
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our comprehensive experimental study explores the effects of the propagation and normalization operators, and the number of steps in the RFP trajectory, on both node-level and graph-level tasks. The results indicate that RFP improves the performance of MPNNs, with or without spectral and random encoding as well as other recent methods. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science, Ben-Gurion University of the Negev, Israel. 2Department of Computing, Imperial College London, UK. 3Department of Computer Science, Purdue University, USA. 4Department of Computer Science, Bar-Ilan University, Israel. 5NVIDIA Research. |
| Pseudocode | Yes | The pseudocode of our method can be found in Appendix D. |
| Open Source Code | No | The paper mentions using PyTorch and PyTorch-Geometric for implementation but does not provide an explicit statement about releasing its source code for the described methodology or a link to a repository. |
| Open Datasets | Yes | We consider both the OGBG-MOLHIV (Hu* et al., 2020) and the ZINC-12k (500k budget) (Sterling & Irwin, 2015; Gómez-Bombarelli et al., 2018; Dwivedi et al., 2020) molecular datasets, where the tasks consist of predicting whether an input molecule inhibits HIV replication in the former, and the value of the constrained solubility in the latter. |
| Dataset Splits | Yes | In all datasets, we use the standard 10 splits from Pei et al. (2020). |
| Hardware Specification | Yes | Our code is implemented using PyTorch (Paszke et al., 2019) and PyTorch-Geometric (Fey & Lenssen, 2019), and all our experiments are run on Nvidia RTX3090 GPUs with 24GB of memory. |
| Software Dependencies | No | The paper mentions using “PyTorch (Paszke et al., 2019) and PyTorch-Geometric (Fey & Lenssen, 2019)” but does not provide specific version numbers for these software dependencies (e.g., PyTorch 1.9). |
| Experiment Setup | Yes | The hyperparameters values were determined using a grid search, and the considered values for those hyperparameters are reported in the following: lr {1e 4, 1e 3, 1e 2, 1e 1}, wd {0, 1e 6, 1e 5, 1e 4, 1e 3}, drop {0, 0.5}, L {2, 4, 8, 16}, c {64, 128, 256}, P {2, 4, 8, 16, 32}, k {4, 8, 16, 32, 64}, B {5, 10}. |