Listing projects that use MatlabAutoDiff

[martinResearch]

Please add a comment in this issue that describes you projet that uses MatlabAutoDiff.

[noblec04]

https://github.com/noblec04/MatlabGP - A GP and NN library that is using AutoDiff for HP finding, for providing gradients for BO adaptive sampling.

I've added/modified a few functions, like an SVD gradient (though with loops through dimensions,, so not super performant), FFTs, modified mldivide for nY>1 and added some special cases for scalar values for gradients through ODE solvers. I hope this is ok.

[martinResearch]

@noblec04 would like to submit a PR with your changes?

[uoa1184615]

Have used a revised version of AutoDiff to empower, specifically computation of Jacobians for the Equation-Free Patch scheme that is designed for multiscale simulation and analysis. AutoDiff may well have other uses in this Toolbox. It includes a vectorized and sparse-enabled adaptation of Chris Noble's FFT. See the repository https://github.com/uoa1184615/EquationFreeGit

[uoa1184615]

My repository extension of AutoDiff now contains a vectorized and sparse-enabled adaption of Chris Noble's svd()---what fun! I also copied in the other extra functions Chris's copy of AutoDiff possesses. See https://github.com/uoa1184615/EquationFreeGit

[martinResearch]

My repository extension of AutoDiff now contains a vectorized and sparse-enabled adaption of Chris Noble's svd()---what fun! I also copied in the other extra functions Chris's copy of AutoDiff possesses. See https://github.com/uoa1184615/EquationFreeGit

Great :) please feel free to submit a pool request with some of these additions.

[uoa1184615]

Hi Martin,

because I downloaded the package, not cloned the repository, I do not think I can put in a pull request through github. Attached is the two files I changed, plus one additional info file. Most of my edits are flagged by my initials "AJR". Use as desired.

BTW I also rewrapped some of the chunks of comments to a maximum linelength of 60 characters. It is well-established that longer lines are harder to comprehend, and a maximum line-length of 50-60 characters is recommended.

Tony

PS I forgot to mention that my extension to the SVD also handles mxn matrices, whereas Chris's is restricted to square. Outstanding task is to extend to cover general complex matrices---it seems to require some extra terms but not yet clear.

Archive.zip

[uoa1184615]

Hi Martin, could not relax until I had sorted out AD of SVD of complex matrices, including non-square. A little tricky because SVD factorisation is not an analytic function of the matrix, so its differentials have to be used with care.

Archive.zip

Anyway, the attached revised AutoDiff.svd() works for all test cases I have thrown at it: see by executing attached script diffSVD.m This script is self documenting and is explained in attached diffSVD.pdf

Tony

[martinResearch]

Hi @uoa1184615 , Thank you very much for the code, you did an amazing work. I would not have thought computing derivatives of the SVD was such a hard problem! Before adding it to the MatlabAutoDiff package I would like to have tests I can run to check the results are valid that I can add to MatlabAutoDiff/src/AutoDiff_tests.m at master · martinResearch/MatlabAutoDiff <https://github.com/martinResearch/MatlabAutoDiff/blob/master/src/AutoDiff_tests.m> Also it would be great if you could fork the repo, and create a PR. You would appear as a contributor to the repo in the git history and it would make it easier to discuss details through comments. Best regards Martin

…

On Mon, 22 Jun 2026 at 01:57, Tony Roberts ***@***.***> wrote: Hi Martin, could not relax until I had sorted out AD of SVD of complex matrices, including non-square. A little tricky because SVD factorisation is not an analytic function of the matrix, so its differentials have to be used with care. Archive.zip <https://github.com/user-attachments/files/29186396/Archive.zip> Anyway, the attached revised AutoDiff.svd() works for all test cases I have thrown at it: see by executing attached script diffSVD.m This script is self documenting and is explained in attached diffSVD.pdf Tony — Reply to this email directly, view it on GitHub <#16 (comment)> or unsubscribe <https://github.com/notifications/unsubscribe-auth/AELQGRW7WUBQ3V7MTUMEWZD5BB757BFKMF2HI4TJMJ2XIZLTSSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDUOJ2WLJDOMFWWLLTXMF2GG2C7MFRXI2LWNF2HTAVFOZQWY5LFUVUXG43VMWSG4YLNMWVXI2DSMVQWIX3UPFYGLAVFOZQWY5LFU4YTEMRRGI3TFJDOMFWWLKDBMN2G64S7NFSKY43VMJVGKY3UL52HS4DFVREXG43VMVBW63LNMVXHJJTUN5YGSY3TSKBKI5DZOBS2U4TFOBXXG2LUN5ZHTJLWMFWHKZNIGY4DAOJTGA2TLAVEOR4XAZNFNFZXG5LFUV3GC3DVMWVDCNJWGE2DCMZZGY4KO5DSNFTWOZLSUZRXEZLBORSQ> . You are receiving this email because you authored the thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .

[martinResearch]

@uoa1184615 thinking more about it, I wonder if expose AutoDiff objects for U and V is not risky in the sense it can encourage incorrect use of the the derivatives. I created this issue #22 to discuss possible remediation.

[uoa1184615]

HI Martin, a PR is beyond my straightforward use of git, github, and github-Desktop so far.

However, I have I have prepared revised clone of your MatlabAutoDiff repository. With two commits ready to be pushed somewhere, somehow. But the committed revisions are currently on the "master" branch.

How can I use git, github, or github-Desktop to make a PR for you?

Tony