Improvements on Uncertainty Quantification for Node Classification via Distance Based Regularization

Authors: Russell Hart, Linlin Yu, Yifei Lou, Feng Chen

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive comparison experiments on eight standard datasets and demonstrate that the proposed regularization outperforms the state-of-the-art in both OOD detection and misclassification detection.
Researcher Affiliation Academia Russell Alan Hart The University of Texas at Dallas rah150030@utdallas.edu Linlin Yu The University of Texas at Dallas linlin.yu@utdallas.edu Yifei Lou University of North Carolina at Chapel Hill yflou@unc.edu Feng Chen The University of Texas at Dallas feng.chen@utdallas.edu
Pseudocode No The paper describes algorithms and models mathematically and textually but does not include a distinct pseudocode block or an algorithm labeled as such.
Open Source Code Yes The code is available at https://github.com/neoques/Graph-Posterior-Network.
Open Datasets Yes Datasets We use three citation networks (i.e. Cora ML, Cite Seer, Pubmed) [4], two co-purchase datasets [31] (i.e. Amazon Computers, Amazon Photos), two coauthor datasets [31] (i.e. Coauthor CS and Coauthor Physics) and a large dataset OGBN Arxiv [16].
Dataset Splits Yes We use the same train/val/test split of 5/15/80 as [32].
Hardware Specification No The paper does not specify the hardware used for experiments, such as CPU or GPU models, or cloud instance types.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., specific PyTorch or Python versions).
Experiment Setup Yes Besides, hyperparameters that we tune include entropy regularization weight, distance-based regularization format (whether RD or Rα), and weighting parameters (λ1, λ2), which are optimized based on the validation cross-entropy for each specific dataset. For a comprehensive overview of the hyperparameter configuration and ablation study, please refer to Appendix D. We use the Adam optimizer with a learning rate of 0.01.