Self-supervised Auxiliary Learning with Meta-paths for Heterogeneous Graphs
Authors: Dasol Hwang, Jinyoung Park, Sunyoung Kwon, KyungMin Kim, Jung-Woo Ha, Hyunwoo J. Kim
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experiments demonstrate that the proposed method consistently improves the performance of link prediction and node classification on heterogeneous graphs. ... 4 Experiments |
| Researcher Affiliation | Collaboration | Dasol Hwang1 , Jinyoung Park1 , Sunyoung Kwon4 Kyung-Min Kim2,3 , Jung-Woo Ha2,3 , Hyunwoo J. Kim1 Korea University1, NAVER AI LAB2, NAVER CLOVA3, Pusan National University4 {dd_sol, lpmn678, hyunwoojkim}@korea.ac.kr skwon@pusan.ac.kr, {kyungmin.kim.ml, jungwoo.ha}@navercorp.com |
| Pseudocode | Yes | Algorithm 1 Self-supervised Auxiliary Learning |
| Open Source Code | Yes | Our code is publicly available at https://github.com/mlvlab/SELAR. |
| Open Datasets | Yes | We use two public benchmark datasets from different domains for link prediction: Music dataset Last-FM and Book dataset Book-Crossing, released by KGNN-LS [52], Ripple Net [53]. We use two datasets for node classification: citation network datasets ACM and Movie dataset IMDB, used by HAN [46] for node classification tasks. |
| Dataset Splits | Yes | Algorithm 1 Self-supervised Auxiliary Learning Input: training data for primary/auxiliary tasks Dpr, Dau, mini-batch size Npr, Nau ... Dpr(train) m , Dpr(meta) m CVSplit(Dpr m , c) Split Data for CV ... We used 3-fold cross validation and the gradients of Θ w.r.t different meta-datasets are averaged to update Θk, see Algorithm 1. |
| Hardware Specification | No | Our experiments were mainly performed based on NAVER Smart Machine Learning platform (NSML) [54, 55]. This refers to a platform, not specific hardware components like GPU or CPU models. |
| Software Dependencies | No | The paper mentions |
| Experiment Setup | No | Implementation details are in the supplement. The main text does not provide specific hyperparameter values or detailed system-level training settings. |