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..
Harnessing Feature Resonance under Arbitrary Target Alignment for Out-of-Distribution Node Detection
Authors: Shenzhi Yang, Junbo Zhao, Sharon Li, Shouqing Yang, Dingyu Yang, Xiaofang Zhang, Haobo Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on a total of thirteen real-world graph datasets empirically demonstrate that RSL achieves state-of-the-art performance. |
| Researcher Affiliation | Academia | 1 Zhejiang University 2 Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security 3 Department of Computer Sciences, University of Wisconsin-Madison 4 School of Computer Science and Technology, Soochow University 5 Innovation and Management Center, School of Software Technology(Ningbo), Zhejiang University Corresponding to: EMAIL |
| Pseudocode | Yes | H Algorithm Pseudo-code Algorithm 1 Resonance-based Separate and Learn (RSL) Framework for Category-Free OOD Detection |
| Open Source Code | Yes | The code is available via https://github.com/Shenzhi Yang2000/RSL. |
| Open Datasets | Yes | Datasets. We conduct extensive experiments to evaluate RSL on a total of nine real-world OOD node detection datasets: six multi-category datasets, Squirrel [Rozemberczki et al., 2021], Wiki CS [Mernyei and Cangea, 2020], Cora, Citeseer, Pubmed [Kipf and Welling, 2016a], and Chameleon [Rozemberczki et al., 2021] and three binary classification fraud detection datasets: Yelp Chi [Rayana and Akoglu, 2015], Amazon [Mc Auley and Leskovec, 2013], and Reddit [Kumar et al., 2019]. The statistics of these datasets are summarized in Table 1. Additionally, we validate our method on four graph-level OOD detection datasets, including ENZYMES, PROTEINS [Morris et al., 2020], Clin Tox, and LIPO [Wu et al., 2018]. We provide detailed dataset description in the Appendix E.3. |
| Dataset Splits | Yes | We allocate 40% of the ID class nodes for training, with the remaining nodes split into a 1:2 ratio for validation and testing, ensuring stratified random sampling based on ID/OOD labels. ... We follow the data processing procedure used in GOOD-D [Liu et al., 2023] that 90% of ID samples are used for training, and 10% of ID samples and the same number of OOD samples are integrated together for testing. |
| Hardware Specification | Yes | Our implementation is based on Ubuntu 20.04, Cuda 12.1, Pytorch 2.1.2, and Pytorch Geometric 2.6.1. All the experiments run with an NVIDIA 3090 with 24GB memory. |
| Software Dependencies | Yes | Our implementation is based on Ubuntu 20.04, Cuda 12.1, Pytorch 2.1.2, and Pytorch Geometric 2.6.1. |
| Experiment Setup | Yes | As shown in Table 10. |