Good morning,
there is a debate on issues of bias in classic approaches to representational similarity analysis. On this, see
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006299
https://arxiv.org/abs/2102.08931
An SPM toolbox designed with this issue in mind (for use in the searchlight approach) is publicly available here:
https://github.com/roberto-viviani/rsa-rsm
Best wishes,
Roberto Viviani
On Wednesday, January 26, 2022 16:57 CET, Guillaume Flandin <[log in to unmask]> wrote:
> Hi All,
>
> Thanks for the great comments here. Martin, I presume this is the link
> to the toolbox you are referring to at the end of your message?
>
> https://github.com/rsagroup/rsatoolbox
>
> I will also add to the mix Variational RSA for another perspective on
> the topic from an SPM viewpoint:
>
> https://www.sciencedirect.com/science/article/pii/S1053811919305610
>
> Best regards,
> Guillaume.
>
>
> On 26/01/2022 15:48, Martin Hebart wrote:
> > ⚠ Caution: External sender
> >
> >
> > Hi Maxime,
> >
> > I'm an author of TDT and even though I'm really happy if you use it (and
> > thanks Tibor for the kind words!), I think there are definitely things
> > that need to be improved for RSA for version 4.0 of the toolbox. For
> > getting RDMs from brain data, it should really be quite straightforward
> > with TDT, specifically for crossnobis distance. Also, running basic RSA
> > should work out of the box (e.g. getting the correlation of a model RDM
> > with brain RDMs). The devil lies in the detail though: do you want to
> > average data of the same conditions, do you want to do cross-validation,
> > what distance metric, what comparison, multiple regression, do you want
> > to use all cells of the matrix or only some, etc. etc. People on the
> > support forum are seem to be most confused about RSA, and it's probably
> > because there are so many ways to do it and our toolbox has adopted
> > these methods over time. I doubt it would be a lot easier with other
> > Matlab-based tools but I just wanted to mention it might take a bit of
> > time. I try answering questions in the forum though. :)
> >
> > If you are firm in Python and like object oriented stuff, it might also
> > be worth looking into the new RSA toolbox.
> >
> > Best,
> > Martin
> >
> > Am Mi., 26. Jan. 2022 um 16:15 Uhr schrieb Auer, Tibor
> > <[log in to unmask] <mailto:[log in to unmask]>>:
> >
> > Hi Maxime,
> >
> > The Cambridge tool is solid, but I would hesitate to pick a new tool
> > without detailed documentation and support.
> >
> > I have previously used CoSMoMVPA (https://www.cosmomvpa.org
> > <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.cosmomvpa.org%2F&data=04%7C01%7Cg.flandin%40UCL.AC.UK%7Cdb1638df917241873f7408d9e0e34de5%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637788089137886653%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=brDCJm8%2FpJjFBsqp57qcH%2B%2BgdvRa%2FO14mknl4gAjo7U%3D&reserved=0>),
> > which is easy to use and have an active user group. I have also
> > integrated it in Automatic Analysis
> > (https://github.com/automaticanalysis/automaticanalysis) to
> > facilitate adoption, and I also have an example pipeline for RSA ,
> > which I have not published yet.
> >
> > I have recently used TDT, which seems the best and most solid of
> > all, and an active community is certainly a plus. I have also
> > integrated TDT in Automatic Analysis for all sorts of inputs
> > (voxel-based, ROI-based, connectivity); however, I have not used it
> > for RSA, yet, but it seems straightforward to do it.
> >
> > Kind regards,
> > Tibor
> >
> > Auer, Tibor M.D. Ph.D.
> > Research Fellow
> > School of Psychology, Faculty of Health and Medical Sciences
> > University of Surrey, Guildford GU2 7XH
> > [log in to unmask] <mailto:[log in to unmask]>
> > @TiborAuer
> >
> > -----Original Message-----
> > From: SPM (Statistical Parametric Mapping) <[log in to unmask]
> > <mailto:[log in to unmask]>> On Behalf Of Maxime Carri ere
> > Sent: Wednesday, January 26, 2022 12:46 PM
> > To: [log in to unmask] <mailto:[log in to unmask]>
> > Subject: [SPM] Representational Similarity Analysis || Where to start?
> >
> > Hello everyone,
> >
> >
> >
> > I have just started my PhD and I have been asked to do
> > Representational Similarity Analysis with fMRI. However, there is no
> > one in the lab who is working on this technique.
> >
> > I would like to ask you if you know any good ressources and/or
> > package for this purpose? I already selected a couple and your
> > opinion would be greatly appreciated!
> >
> > (1) The Decoding Toolbox
> > https://sites.google.com/site/tdtdecodingtoolbox/
> > <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsites.google.com%2Fsite%2Ftdtdecodingtoolbox%2F&data=04%7C01%7Cg.flandin%40UCL.AC.UK%7Cdb1638df917241873f7408d9e0e34de5%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637788089137886653%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=XgSkbLQP25bkgaYIbeOxk9iO0FAGbHvf%2Bzhzzz2as00%3D&reserved=0>
> > With a relatively active community on https://neurostars.org/
> > <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fneurostars.org%2F&data=04%7C01%7Cg.flandin%40UCL.AC.UK%7Cdb1638df917241873f7408d9e0e34de5%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637788089137886653%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=wpjk5qb5uT%2F96tfJoagYiy7RcuIz7Iwnff3atywuFGA%3D&reserved=0>
> >
> > (2) The Cambridge Toolbox
> > https://www.mrc-cbu.cam.ac.uk/methods-and-resources/toolboxes/
> > <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.mrc-cbu.cam.ac.uk%2Fmethods-and-resources%2Ftoolboxes%2F&data=04%7C01%7Cg.flandin%40UCL.AC.UK%7Cdb1638df917241873f7408d9e0e34de5%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637788089137886653%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=T73kdJyoBzLnJn7EYcpOsKJw%2FCT2AUbY3gr6BLvBHwQ%3D&reserved=0>
> > However, I could not find any active community.
> >
> > If you have any other Toolbox, package, or ressources please share ;)
> >
> > Have a nice day,
> >
> > Maxime
> >
>
> --
> Guillaume Flandin, PhD
> Wellcome Centre for Human Neuroimaging
> UCL Queen Square Institute of Neurology
> London WC1N 3BG
|