notesum.ai

Published at November 13

SANDWICH: Towards an Offline, Differentiable, Fully-Trainable Wireless Neural Ray-Tracing Surrogate

cs.NI

cs.AI

Released Date: November 13, 2024

Authors: Yifei Jin¹, Ali Maatouk², Sarunas Girdzijauskas³, Shugong Xu⁴, Leandros Tassiulas², Rex Ying²

Aff.: ¹KTH Royal Institute of Tech. & Ericsson AB, Stockholm, Sweden; ²Yale University, New Haven, USA; ³KTH Royal Institute of Tech., Stockholm, Sweden; ⁴Shanghai University, Shanghai, China

Arxiv: http://arxiv.org/abs/2411.08767v1

Metric (\unit)	MAE(Model, GT)			MAE(GCE, Model)			KL_div(pdf(GCE)-pdf(Model))
On Test set:	Checkerboard	Genz	GenDiag	Checkerboard	Genz	GenDiag	Checkerboard	Genz	GenDiag
GCE (Topline)	3.422	3.181	3.039	0	0	0	0	0	0
SANDWICH	3.966	3.804	3.689	1.494	1.64	1.422	0.041	0.0534	0.0317
MLP(Reproduced)	5.337	5.846	4.729	3.760	4.291	3.496	0.2827	0.6003	0.1913
KNN(Reproduced) 1	11.21	11.04	4.412	9.856	9.939	4.963	0.1447	0.1565	0.1468
KNN(Reproduced) 2	10.71	10.59	4.180	9.446	9.542	4.822	0.1431	0.1516	0.1508