notesum.ai

Published at October 21

SeaDAG: Semi-autoregressive Diffusion for Conditional Directed Acyclic Graph Generation

cs.CV

cs.AI

cs.MM

Released Date: October 21, 2024

Authors: Xinyi Zhou, Xing Li, Yingzhao Lian, Yiwen Wang, Lei Chen, Mingxuan Yuan, Jianye Hao, Guangyong Chen, Pheng Ann Heng

Arxiv: https://arxiv.org/abs/2410.16119v1

Refer to caption

Generation

Mode

Model

Validity

\uparrow

NSPDK

\downarrow

FCD

\downarrow

Unique

\uparrow

Novelty

Unconditional

GraphAF

74.43

0.020

5.27

88.64

86.59

GraphDF

93.88

0.064

10.93

98.58

98.54

MoFlow

91.36

0.017

4.47

\ul98.65

94.72

EDP-GNN

47.52

0.005

2.68

99.25

86.58

GraphEBM

8.22

0.030

6.14

97.90

97.01

SPECTRE

87.3

0.163

47.96

35.70

97.28

GDSS

95.72

0.003

2.90

98.46

86.27

GRAPHARM

90.25

\ul0.002

\ul1.22

95.62

70.39

LDM-3DG

100.0

0.009

2.44

97.57

89.89

DiGress

\ul99.00

0.0005

0.36

96.66

33.40

Conditional

LDM-3DG

100.0

0.018

4.72

81.09

92.03

DiGress

\ul99.28

0.0004

\ul0.72

96.34

47.21

SeaDAG

100.0

\ul0.002

0.36

\ul93.01

55.15