Learning Substructure Invariance for Out-of-Distribution Molecular Representations
Authors: Nianzu Yang, Kaipeng Zeng, Qitian Wu, Xiaosong Jia, Junchi Yan
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on ten real-world datasets demonstrate that our model has a stronger generalization ability than existing methods under various out-of-distribution (OOD) settings, despite the absence of manual specifications of environments. |
| Researcher Affiliation | Academia | Nianzu Yang, Kaipeng Zeng, Qitian Wu, Xiaosong Jia, Junchi Yan Department of Computer Science and Engineering Mo E Key Lab of Artificial Intelligence Shanghai Jiao Tong University {yangnianzu,zengkaipeng,echo740,jiaxiaosong,yanjunchi}@sjtu.edu.cn Junchi Yan is the correspondence author who is also with Shanghai AI Laboratory. |
| Pseudocode | Yes | Algorithm 1: The training procedure. |
| Open Source Code | Yes | Our source code is publicly available at https://github.com/yangnianzu0515/Mole OOD. |
| Open Datasets | Yes | The four datasets BACE, BBBP, SIDER and HIV, are from by Open Graph Benchmark (OGB) [23]. ... The other six datasets are generated by the dataset curator provided by Drug OOD [27]. |
| Dataset Splits | Yes | We use the default train/val/test split with ratio 8:1:1. |
| Hardware Specification | Yes | Experiments are performed on 10 benchmark datasets and repeated 5 times with mean and standard deviation reported, running on a machine with i9-10920X CPU, RTX 3090 GPU and 128G RAM. |
| Software Dependencies | No | The paper mentions that training details are specified in Appendix C, but the provided text does not contain explicit software names with version numbers for reproducibility. |
| Experiment Setup | Yes | We use the default train/val/test split with ratio 8:1:1. ... Each of the method is configured using the same parameters reported in the original paper or selected by grid search. For the sake of fairness, the embedding size of all methods are set to be equal in comparison. We specify the training details in the Appendix C. |