notesum.ai

Published at October 20

LAC: Graph Contrastive Learning with Learnable Augmentation in Continuous Space

cs.AI

cs.CY

cs.HC

cs.LG

Released Date: October 20, 2024

Authors: Zhenyu Lin¹, Hongzheng Li¹, Yingxia Shao¹, Guanhua Ye¹, Yawen Li¹, Quanqing Xu¹

Aff.: ¹School of Computer Science, Beijing University of Posts and Telecommunications

Arxiv: https://arxiv.org/abs/2410.15355v1

Refer to caption

Category	Model	Cora	CiteSeer	PubMed	CS	Phy	Photo	Computers
Graph Represent	RawFeat* [19]	$56.89\pm 0.08$	$60.70\pm 0.04$	$83.84\pm 0.10$	$73.91\pm 0.20$	$94.53\pm 0.34$	$83.52\pm 0.31$	$78.53\pm 0.00$
Learning	Node2Vec* [43]	$64.50\pm 0.51$	$53.05\pm 1.02$	$74.29\pm 0.39$	$78.03\pm 0.39$	$94.66\pm 0.15$	$89.61\pm 0.24$	$84.39\pm 0.08$
	DGI* [44]	$83.25\pm 0.68$	$72.03\pm 0.65$	$84.75\pm 0.19$	$92.75\pm 0.10$	$94.42\pm 0.67$	$90.76\pm 0.29$	$83.95\pm 0.47$
Graph Generative	GAE* [45]	$76.90\pm 0.25$	$60.60\pm 0.14$	$82.94\pm 0.31$	$90.01\pm 0.71$	$94.92\pm 0.07$	$91.62\pm 0.13$	$85.27\pm 0.19$
Learning	VGAE* [45]	$78.92\pm 0.46$	$61.22\pm 0.11$	$83.07\pm 0.29$	$92.11\pm 0.09$	$94.52\pm 0.01$	$92.20\pm 0.11$	$86.37\pm 0.21$
Manual	GRACE* [19]	$83.32\pm 0.41$	$72.10\pm 0.53$	$85.51\pm 0.37$	$91.12\pm 0.20$	$95.41\pm 0.13$	$92.15\pm 0.25$	$87.25\pm 0.25$
GCL	GCA* [18]	$82.89\pm 0.21$	$72.89\pm 0.13$	$85.12\pm 0.23$	$93.10\pm 0.01$	$95.68\pm 0.05$	$92.49\pm 0.09$	$87.85\pm 0.31$
Frameworks	MVGRL* [5]	$83.11\pm 0.12$	$\uline{73.33\pm 0.03}$	$84.27\pm 0.04$	$93.11\pm 0.12$	$95.33\pm 0.03$	$91.74\pm 0.07$	$87.52\pm 0.11$
	BGRL* [17]	$82.83\pm 1.61$	$72.32\pm 0.89$	$\uline{86.03\pm 0.33}$	$93.31\pm 0.13$	$\uline{95.73\pm 0.05}$	$93.17\pm 0.30$	$88.54\pm 0.03$
	COSTA †[42]	$84.32\pm 0.22$	$72.92\pm 0.31$	$86.01\pm 0.19$	$92.56\pm 0.45$	$95.01\pm 0.09$	$92.56\pm 0.42$	$88.32\pm 0.30$
Statistical	CCA-SSG* [46]	$84.20\pm 0.40$	$73.10\pm 0.30$	$85.39\pm 0.51$	$93.31\pm 0.22$	$95.38\pm 0.06$	$93.14\pm 0.14$	$88.74\pm 0.28$
GCL Frameworks	MC-DCD †[30]	-	-	-	$93.60\pm 0.08$	$95.50\pm 0.04$	$93.31\pm 0.13$	$88.78\pm 0.25$
Automated	JOAO-v2 † [6]	$82.47\pm 0.43$	$70.29\pm 0.44$	$83.81\pm 0.05$	$91.50\pm 0.29$	$94.79\pm 0.53$	$91.39\pm 0.25$	$86.73\pm 0.44$
GCL	ADGCL † [22]	$83.51\pm 0.63$	$72.42\pm 0.36$	$85.45\pm 0.33$	$93.26\pm 0.30$	$95.57\pm 0.09$	$91.45\pm 0.12$	$86.03\pm 0.21$
Frameworks	AutoGCL† [3]	$83.86\pm 0.25$	$72.62\pm 0.50$	$84.27\pm 0.52$	$92.37\pm 0.24$	$95.15\pm 0.14$	$92.25\pm 0.30$	$87.11\pm 0.84$
Spectral Framework	GCL-SPAN† [28]	$\uline{85.51\pm 0.61}$	$72.46\pm 0.35$	$85.45\pm 0.16$	$\uline{93.76\pm 0.38}$	OOM	$92.58\pm 0.17$	$\uline{89.96\pm 0.13}$
Ours	LAC	$\boldsymbol{86.02}\pm\boldsymbol{0.45}$	$\boldsymbol{73.51}\pm\boldsymbol{0.62}$	$\boldsymbol{86.05}\pm\boldsymbol{0.21}$	$\boldsymbol{93.84}\pm\boldsymbol{0.30}$	$\boldsymbol{96.08}\pm\boldsymbol{0.15}$	$\boldsymbol{93.75}\pm\boldsymbol{0.43}$	$\boldsymbol{90.55}\pm\boldsymbol{0.16}$