Fixed Point Theory and Algorithms for Sciences and Engineering

Table 1 NMF results for RRR on the LEDM instances compared with coordinate descent (CD) and the leading heuristic from Vandaele et al. [19]. The RRR results are based on 20 runs for each instance

From: Learning without loss

m	\(r_{+}\)	RRR					CD		heuristic¹
m	\(r_{+}\)	β	ω	rate	sec	GWMs	rate	sec	rate	sec
6	5	0.3	0.6	100%	2	0.02	1.6%	100	100%	19
8	6	0.3	0.7	100%	59	0.61	<1%	—	99%	64
12	7	0.3	0.8	100%	567	6.27	<1%	—	69%	37
16	8	0.3	0.9	100%	3677	41.52	<1%	—	48%	104

¹ Vandaele et al.

Back to article page