notesum.ai

Published at November 15

DuSEGO: Dual Second-order Equivariant Graph Ordinary Differential Equation

cs.LG

cs.AI

Released Date: November 15, 2024

Authors: Yingxu Wang¹, Nan Yin², Mingyan Xiao³, Xinhao Yi⁴, Siwei Liu⁵, Shangsong Liang⁶

Aff.: ¹Mohamed bin Zayed University of Artificial Intelligence, UAE; ²The Hong Kong University of Science and Technology, Hong Kong, China; ³California State Polytechnic University, Pomona, United States; ⁴School of Computing Sciences, University of Glasgow, Scotland; ⁵School of Natural and Computing Sciences, University of Aberdeen, Scotland; ⁶Emirates Aviation University, UAE

Arxiv: http://arxiv.org/abs/2411.10000v1

Methods	MSE			Forward Time/ms
Methods	1000 ts	1500 ts	2000 ts	Forward Time/ms
Linear	6.831 $\pm$ 0.016	20.012 $\pm$ 0.029	39.513 $\pm$ 0.061	0.1
SE(3) Trans	2.483 $\pm$ 0.099	18.891 $\pm$ 0.287	36.730 $\pm$ 0.381	114.6
TFN	1.544 $\pm$ 0.231	11.116 $\pm$ 2.825	23.823 $\pm$ 3.048	27.2
GNN	1.077 $\pm$ 0.004	5.059 $\pm$ 0.250	10.591 $\pm$ 0.352	2.1
Radial Field	1.060 $\pm$ 0.007	12.514 $\pm$ 0.089	26.388 $\pm$ 0.331	3.0
EGNN	0.716 $\pm$ 0.029	2.201 $\pm$ 0.081	4.049 $\pm$ 0.103	5.5
SEGNN	0.481 $\pm$ 0.016	1.552 $\pm$ 0.061	3.294 $\pm$ 0.095	24.5
GraphCon	1.015 $\pm$ 0.011	3.148 $\pm$ 0.091	6.133 $\pm$ 0.143	10.8
SEGNO	0.976 $\pm$ 0.012	2.994 $\pm$ 0.059	7.453 $\pm$ 0.181	21.2
GMN	0.701 $\pm$ 0.018	1.956 $\pm$ 0.096	3.939 $\pm$ 0.151	6.2
DuSEGO-EGNN	0.674 $\pm$ 0.014	1.912 $\pm$ 0.055	3.876 $\pm$ 0.079	13.9
DuSEGO-SEGNN	0.449 $\pm$ 0.012	1.427 $\pm$ 0.047	3.122 $\pm$ 0.081	45.7