. Author manuscript; available in PMC: 2014 Jan 13.

Published in final edited form as: Inf inference. 2012 Dec;1(1):21. doi: 10.1093/imaiai/ias002

Table 6.

Running times (in seconds) of the 3D-ASAP algorithm for the BRIDGE-DONUT graph with n=500 nodes, η =0%, 35%, deg =18, 21, n=533, 541 patches, and average patch sizes 17.8, 20.3. For η =0% we embed the patches using FULL-SDP, while for η =35% we use SNL-SDP since the regularization term improves the localization. For η =35%, during the least-squares step we also use the scaling heuristic, and run the gradient descent algorithm several times, hence the increase in the running time from 6 to 21 s

	η =0%	η =35%
Break G into patches	59	90
Finding WUL subgraphs	210	233
Embedding FULL-SDP	154	(252)
Embedding SNL-SDP	(966)	1368
Denoising patches	30	39
Step 1 (ℤ₂ × SO(3))	15	19
Step 2 (least squares)	6	21
3D-ASAP	474	1770
DISCO	196	197
3D-SP-ASAP	108	186