I just came across your Matrix Factorization Jungle website, and thought you might like to list the Incremented Rank PowerFactorization algorithm that Diego Hernando and I published about a while back:(Rank-Constrained Solutions to Linear Matrix Equations Using PowerFactorization )It's a fast greedy algorithm that solves a rank-constrained version of the matrix recovery inverse problem, and empirically seems to have advantages over convex-relaxation approaches in a number of settings (previously mentioned on Nuit-Blanche:). It's also been used for sparsely-sampled medical imaging reconstruction problems: See:
Here is my response
Justin,I am torn and you might help me in figuring this thing out.You may have noticed that the Matrix Factorization Jungle only features implementations that people can download. It is just too difficult to keep track of any and all implementations that are eventually never available (for whatever reason). I would love to feature your solver. Is there anyway you could make something available even if it is at a prototype level (for a small dataset) ? I could then point to you for people who would want something with more strength.Cheers,Igor.
If you recall, I had a small rant about this a while back in Nobody Cares About You and Your Algorithm and the argument still stands. Help me help you become a rock star. Some people might not take the message kindly as it goes against some old habit of academia but Justin was not one of these people, he wrote back the next day the following:
Hi Igor,With your encouragement, I've created a simple stripped-down Matlab implementation that people can download from here:Cheers,--Justin
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.