High dimensional, tabular deep learning with an auxiliary knowledge graph

Authors: Camilo Ruiz, Hongyu Ren, Kexin Huang, Jure Leskovec

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

Reproducibility Variable Result LLM Response
Research Type Experimental Across 6 d n datasets, PLATO outperforms 13 state-of-the-art baselines by up to 10.19%. We evaluate PLATO against 13 baselines on 10 tabular datasets (6 with d n, 4 with d n).
Researcher Affiliation Academia 1Department of Computer Science, Stanford University 2Department of Bioengineering, Stanford University
Pseudocode Yes PLATO is outlined in Algorithm 1.
Open Source Code Yes Code, data, and the KG are available at https://github.com/snap-stanford/plato.
Open Datasets Yes We use 6 tabular d n datasets, 4 d n datasets [16, 17, 30, 79], and a KG from prior studies [44, 35, 38, 56, 63, 74, 75] (Appendix G, H)... Data is available at https://github.com/snap-stanford/plato.
Dataset Splits Yes We split data with a 60/20/20 training, validation, test split.
Hardware Specification Yes Each model is run on a Ge Force RTX 2080 TI GPU.
Software Dependencies No The paper does not explicitly list software dependencies with version numbers (e.g., Python, specific libraries like PyTorch or TensorFlow with their versions).
Experiment Setup Yes We conduct a random search with 500 configurations of every model (including PLATO) on every dataset across a broad range of hyperparameters (Appendix A). Hyperparameter ranges for PLATO are given in Table 7. Hyperparameter ranges for baseline methods are given in Table 8.