tag:blogger.com,1999:blog-6141980.post7593430725903348423..comments2024-03-20T12:28:35.004-05:00Comments on Nuit Blanche: CS: Least Squares vs Basis Pursuit: Multiplicative Noise Effect on the Donoho-Tanner Phase Transition (Part 5)Igorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-6141980.post-40849622643478640922011-03-28T06:12:22.033-05:002011-03-28T06:12:22.033-05:00Bob,
I am glad we are having this discussion aft...Bob, <br /><br />I am glad we are having this discussion after five posts :-)<br /><br />Yes every (well maybe not every you need to see my explanation below) pixel in these graphs represents the result of 20 reconstructions. I need to detail the nitty grit tweaking of the algorithm at the end of the series of post. The line is the border between the regions where most reconstructions fail and where most reconstructions work. IN particular, the line is drawn in light blue when more than 2 reconstructions have failed over 20. The region where most reconstructions work is always below. The color does not matter as much but I will be releasing the code later.<br /><br />'No trick' means that I am not using the [A I] trick used in the Cross and Bouquet approach. So 'No Trick' means I am doing the vanilla reconstruction using whatever solver I am stating I am using...<br /><br />Not all pixels were computed, since I have an old machine, I need to get these computations running fast. I compute the pixels initially from below (k/N = 1/N) all the way to the border. Then when I get to the border, the algo goes to the next m/N and then recompute some pixels below the border all the way up to the next pixels on the border. It is a simple adaptive algorithm but obviously any publication would need to have the full computations done. Right now, I am going the way of the pirates.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-54183295287097794752011-03-28T05:44:28.085-05:002011-03-28T05:44:28.085-05:00Hi Igor,
Five days later, I am still confused by ...Hi Igor,<br /><br />Five days later, I am still confused by what you are showing. What are the colors? What does "No Trick delta ..." mean? What are the axes? Why do you show these results as images and not lines? Are you evaluating recoveries for every x,y pair? If so, how many?Bob et Carlahttps://www.blogger.com/profile/09965549060044467221noreply@blogger.com