Timeline for SVD when only off-diagonal terms are known
Current License: CC BY-SA 3.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 11, 2016 at 4:38 | comment | added | Pushpendre | The question at this point is not well-posed but here are a few "search directions" which lead to potential approaches for solving your problem. One key term you are looking for is "missing values". "missing value svd" on google throws up many papers and even matlab code. The only constraint you have stated is that the matrix is low rank. This suggests that one of the billion different "low rank matrix completion" approaches could work. Another clue is that the matrix has only positive elements. Maybe the diagonals should also be positive? That would suggest "low rank nmf matrix completion". | |
| Apr 10, 2016 at 18:42 | comment | added | Christian Remling | This of course can't work in this generality, because the off-diagonal terms could be almost zero and everything is dominated by the unknown diagonal. | |
| Apr 10, 2016 at 16:25 | comment | added | kjetil b halvorsen | Can you at least give reference to some of the papers you are surfing? | |
| Apr 10, 2016 at 15:34 | comment | added | Ulderique Demoitre | @Kjetil, unfortunately I don't have many other details, I am just surfing the literature about mixture models, and seems that nobody cares about the absence of diagonal terms, so from this my question.. | |
| Apr 10, 2016 at 15:33 | comment | added | Ulderique Demoitre | @Suvrit and Federico, do you have a reference for your comments? I would be great! | |
| Apr 10, 2016 at 14:36 | comment | added | Suvrit | This can be solved as an approximate optimization problem, without truly doing a true SVD. | |
| Apr 10, 2016 at 11:54 | comment | added | Federico Poloni | If you can easily estimate off-diagonals, can you change basis and use this to estimate the diagonal? The opposite (polarization identities) is a well-known trick. | |
| Apr 10, 2016 at 9:56 | comment | added | kjetil b halvorsen | If you could give more details of your real problem, I'm sure we can help! | |
| S Apr 10, 2016 at 9:30 | history | edited | Martin Hairer | CC BY-SA 3.0 |
tidying up, spelling
|
| S Apr 10, 2016 at 9:30 | history | suggested | kjetil b halvorsen | CC BY-SA 3.0 |
tidying up, spelling
|
| Apr 10, 2016 at 9:29 | comment | added | kjetil b halvorsen | In which context does this arise? Any possibility of building some statistical or probability model for the missing diagonal? | |
| Apr 10, 2016 at 9:28 | review | Suggested edits | |||
| S Apr 10, 2016 at 9:30 | |||||
| Apr 10, 2016 at 9:09 | history | asked | Ulderique Demoitre | CC BY-SA 3.0 |