Neural Trees for Learning on Graphs

Authors: Rajat Talak, Siyi Hu, Lisa Peng, Luca Carlone

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the neural tree architecture for semi-supervised node classification in 3D scene graphs and citation networks (Section 7). Our experiments on 3D scene graphs demonstrate that neural trees outperform standard, local message passing GNNs, by a large margin.
Researcher Affiliation Academia Rajat Talak, Siyi Hu, Lisa Peng, and Luca Carlone Laboratory of Information and Decision Systems (LIDS) Massachusetts Institute of Technology {talak, siyi, lisapeng, lcarlone}@mit.edu
Pseudocode Yes Algorithm 1: H-tree
Open Source Code Yes Our code is publically available at https://github.com/MIT-SPARK/neural_tree
Open Datasets Yes We run semi-supervised node classification experiments on Stanford s 3D scene graph dataset [2]. [...] We use the popular citation network datasets [60]
Dataset Splits Yes We randomly select 10% of the nodes for validation and 20% for testing. The hyper-parameters of the two approaches are separately tuned based on the best validation accuracy, while using all 70% of the remaining nodes for training
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9) needed to replicate the experiment.
Experiment Setup No The paper states that "The hyper-parameters of the two approaches are separately tuned based on the best validation accuracy, while using all 70% of the remaining nodes for training; see supplementary material for details." Since the details are deferred to supplementary material and not in the main text, it does not explicitly provide these details in the main paper.