tag:blogger.com,1999:blog-6141980.post3050886512640296457..comments2024-03-20T12:28:35.004-05:00Comments on Nuit Blanche: Sunday Morning Insight: The Map MakersIgorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-6141980.post-39382137536725316462013-11-18T16:56:00.397-06:002013-11-18T16:56:00.397-06:00Thank you Anonymous that was very thorough. As a ...Thank you Anonymous that was very thorough. As a matter of note, the iinitial issue was really about the fact that SL0 got rejected for publication when in fact it would perform better than a few type other solvers . The point being that the phase transition is really the only acid test here as the paper, you mention, shows.<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-41801726006716694372013-11-14T09:26:20.195-06:002013-11-14T09:26:20.195-06:00SL0 is definitely not at the top of the heap w.r.t...SL0 is definitely not at the top of the heap w.r.t phase transitions, but it is one of the better algorithms for the student-t type of signals that come out of wavelet transforms and DCTs and such. For other signal types, however, it can perform quite poorly. <br /><br />In terms of speed, SL0 is pretty fast, but it is not as fast as greedy methods for small problems, nor as fast as first-order algorithms (FISTA, SPGL1, AMP, etc.) for large problems because of its complexity scaling.<br /><br />The preprint http://arxiv.org/abs/1207.3107 has numerical evidence of these claims. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-6141980.post-87393898356851787972013-11-11T19:57:32.773-06:002013-11-11T19:57:32.773-06:00The main difference between AMP and SL0 is that AM...The main difference between AMP and SL0 is that AMP is much less complex (a few matrix-vector multiply) as opposed to an SVD for SL0. The point about SL0 was really was that it had been rejected from publications when in fact it did better than quite a few other solvers (before AMP). The second point was that a more robust version of SL0 against noise did not see the light of the day because of the initial rejection of the SL0 paper. Finally, there was recently a paper on an improvement of SL0 called SL1 or SL0-mod that did improve further the phase transition of the original SL0 although not to the extent of GAMP of Phil Schniter et al. Hope this helps,<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-26764411213047839512013-11-11T09:18:39.939-06:002013-11-11T09:18:39.939-06:00Is SL0 still at the top of the heap of phase trans...Is SL0 still at the top of the heap of phase transitions? Did it hold up under scrutiny? I thought these optimization approaches were supposed to be not as good as the AMP type stuff.Anonymousnoreply@blogger.com