Scalable Deep Generative Relational Model with High-Order Node Dependence
Authors: Xuhui Fan, Bin Li, Caoyuan Li, Scott SIsson, Ling Chen
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate its competitive performance through improved link prediction performance on a range of real-world datasets. In our analyses on a range of real-world relational datasets, we demonstrate that the SDREM can achieve superior performance compared to traditional Bayesian methods for relational data, and perform competitively with other approaches. As the SDREM is the first Bayesian relational model to use neighbourhood-wise propagation to build the deep network architecture, we note that it may straightforwardly integrate other Bayesian methods for modelling high-order node dependencies in relational data, and further improve relationship predictability. |
| Researcher Affiliation | Academia | Xuhui Fan1, Bin Li2, Scott A. Sisson1, Caoyuan Li3, and Ling Chen3 1School of Mathematics & Statistics, University of New South Wales, Sydney 2Shanghai Key Lab of IIP & School of Computer Science, Fudan University 3Faculty of Engineering and IT, University of Technology, Sydney {xuhui.fan, scott.sisson}@unsw.edu.au; libin@fudan.edu.cn |
| Pseudocode | No | Figure 1 presents a 'generative process of a SDREM' with numbered steps (1)-(6). While structured, this describes the generative model mathematically and not as an algorithm's pseudocode. |
| Open Source Code | No | The paper does not provide any concrete access information (e.g., a specific repository link or an explicit statement of code release) for its source code. |
| Open Datasets | Yes | We examine four real-world datasets: three standard citation networks (Citeer, Cora, Pubmed [32] and one protein-to-protein interaction network (PPI) [38]. [32] Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi Rad. Collective classification in network data. In AI magazine, pages 29 93, 2008. [38] Marinka Zitnik and Jure Leskove. Predicting multicellular function through multi-layer tissue networks. In Bioinformatics, pages i190 i198, 2017. |
| Dataset Splits | Yes | Unless specified, reported AUC values are obtained by using 90% (per row) of the data as training data and the remaining 10% as test data. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running its experiments. |
| Software Dependencies | No | The paper discusses various models and techniques (e.g., Gibbs sampling, Ber-Poisson link function, Gamma Belief Networks, Dirichlet Belief Networks) but does not provide specific software names with version numbers that would allow for replication of the experimental environment. |
| Experiment Setup | Yes | For hyper-parameters we specify M Gam(N, 1) for all datasets, and {γ(1) d }d, {γ(l) 1 , γ(l) 0 }l, {c(l)}l are all given Gam(1, 1) priors. Each replicate uses 2000 MCMC iterations with the first 1000 discarded as burn-in. When testing the effect of different values of K we fixed L = 3, and when varying L we fixed K = 20. Considering the computational complexity and modelling power, we set K = 20 and L = 4 for the remaining analyses in this paper. |