Equivariant Graph Hierarchy-Based Neural Networks

Authors: Jiaqi Han, Wenbing Huang, Tingyang Xu, Yu Rong

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Considerable experimental evaluations verify the effectiveness of our EGHN on several applications including multi-object dynamics simulation, motion capture, and protein dynamics modeling.
Researcher Affiliation Collaboration Jiaqi Han1 , Wenbing Huang2,3 , Tingyang Xu4, Yu Rong4 1 Department of Computer Science and Technology, Tsinghua University 2 Gaoling School of Artificial Intelligence, Renmin University of China 3 Beijing Key Laboratory of Big Data Management and Analysis Methods, Beijing, China 4 Tencent AI Lab
Pseudocode No The paper details the mathematical formulations of its components (EMMP, E-Pool, E-Un Pool) using equations, but it does not include any distinct pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/hanjq17/EGHN.
Open Datasets Yes We further evaluate our model on CMU Motion Capture Databse [4]... We adopt Ad K equilibrium trajectory dataset [28] via MDAnalysis toolkit [23].
Dataset Splits Yes we leverage the random split adopted by [14], which includes 200 frame pairs for training, 600 for validation, and another 600 for testing. As for running, we follow a similar strategy and obtain a split with 200/240/240 frame pairs... We split the dataset into train/validation/test sets along the timeline that contain 2481/827/878 frame pairs respectively.
Hardware Specification No The provided text of the paper does not contain specific details about the hardware used, such as GPU or CPU models. Although the paper's checklist indicates details are in the Appendix, this information is not present in the main content provided.
Software Dependencies No The paper mentions using 'MDAnalysis toolkit [23]' but does not provide specific version numbers for this or any other software dependencies, which would be necessary for full reproducibility.
Experiment Setup Yes We assign the node feature as the norm of the velocity vi 2, and the edge attribute as cicj for the edge connecting node i and j, following the setting in [27]... Detailed hyper-parameter settings are in Appendix.