@RossGayler@aus.social
@RossGayler@aus.social avatar

RossGayler

@RossGayler@aus.social

he/him
Independent researcher working on Vector Symbolic Architectures (#VSA), also known as Hyperdimensional Computing (#HDC).

Also here for the:
#Rstats
#QuartoPub
#OpenScience
#reproducibility
#OpenAccess
#rOpenSci
#CogSci
#CompCogSci
#MathPsych

#TwitterMigration

This profile is from a federated server and may be incomplete. Browse more on the original instance.

RossGayler, to mathematics
@RossGayler@aus.social avatar

Maths/CogSci/MathPsych lazyweb: Are there any algebras in which you have subtraction but don't have negative values? Pointers appreciated. I am hoping that the abstract maths might shed some light on a problem in cognitive modelling.

The context is that I am interested in formal models of cognitive representations and I want to represent things (e.g. cats), don't believe that we should be able to represent negated things (i.e. I don't think it should be able to represent anti-cats), but it makes sense to subtract representations (e.g. remove the representation of a cat from the representation of a cat and a dog, leaving only the representation of the dog).

This might also be related to non-negative factorisation: https://en.wikipedia.org/wiki/Non-negative_matrix_factorization

@cogsci

RossGayler,
@RossGayler@aus.social avatar

@Heterokromia @cogsci

Thanks. Modulo arithmetic is actually of interest for other reasons but I think it's not quite what I'm after here.

Using your arithmetic example and assumming rep(cat) = 1 and rep(dog) = 2 I would want behaviours like:

rep(dog and cat) = 2 + 1 = 3
3 - 2 = 1
3 - 1 = 2
2 - 2 = 0
2 - 1 = 2
1 - 2 = 1

I suspect that means that the objects of the algebra have to be multidimensional, rather than unidimensional (as numbers appear to be).

adredish, to random
@adredish@neuromatch.social avatar

It's really unfortunate that reviews are seen as competitive and confrontational. I want my reviews to be suggestions to help the authors.

Because of that, I don't want the reviews I write made public. Reviews are for the editor to assess the author's work and for the author to make the work better. They are not commentary for the public to understand the work.

Fixing the paper is the author's job. Not mine. I want to be able to say "There was a paper a long time ago by someone like XYZ who you should go read." Or "There's a whole literature here by people like ABC to go engage with." Finding that literature is the author's job. Not mine. I want to be able to say "what if X is not linear?" or "what if Y happens?". I don't want to have to work out the math myself. That's the author's job. Not mine. None of this should be public.

Moreover, I need a process to say "I have the following concerns with my own review. I have this knowledge and not that. So you can trust me here, but not there." (Which deanonymizes the review, BTW.) I need a process where I can say to the editor, I'm concerned about A, but it's not my knowledge-base, or I think they should fix B, but it's not that important if they don't. Deciding what of that to communicate to the authors is the editor's job. Not mine.

Given the current ecosystem of publishing reviews and of preventing me from providing context to the editors separate from information sent to the authors, I am very seriously considering turning down all review requests from now on. (But it seems unfair to add publications to the ecosystem while not doing my part to review them.)

😠​ Frustrated.

RossGayler,
@RossGayler@aus.social avatar

@neuralreckoning @knutson_brain @adredish I suspect the extent to which publication is a competition varies by field/venue. I work in a very niche computer science adjacent topic, which is almost universally misunderstood by readers outside the topic area. When reviewing, the vast majority of my comments are aimed at improving understanability, and explicitly marked as to be used at the author's discretion. I only do this for papers I recommend be accepted, and I can't recall one being rejected.

RossGayler,
@RossGayler@aus.social avatar

@neuralreckoning @knutson_brain @adredish Indeed. My attitude to publication peer review is possibly not aligned with any of the other participants. I'd be happy to review preprints and consider reprinting to be publication. (As an independent researcher with no access to an institutional library, I consider anything paywalled to be unpublished and would not cite it if I could possibly avoid it. I think paywalled papers are effectively "personal communication".)

RossGayler, to machinelearning
@RossGayler@aus.social avatar

Most of the Artificial Neural Net simulation research I have seen (say, at venues like NeurIPS) seems to take a very simple conceptual approach to analysis of simulation results - just treat everything as independent observations with fixed effects conditions, when it might be better conceptualised as random effects and repeated measures. Do other people think this? Does anyone have views on whether it would be worthwhile doing more complex analyses and whether the typical publication venues would accept those more complex analyses? Are there any guides to appropriate analyses for simulation results, e.g what to do with the results coming from multi-fold cross-validation (I presume the results are not independent across folds because they share cases).

@cogsci #CogSci #CognitiveScience #MathPsych #MathematicalPsychology #NeuralNetworks #MachineLearning

RossGayler,
@RossGayler@aus.social avatar

@jonny @cogsci
Thanks for the confirmation of the observation. I am asking around elsewhere for an introduction/guide/tutorial on statistical analysis of computational simulation studies.

RossGayler,
@RossGayler@aus.social avatar

@jonny @cogsci I have just updated that post, clarifying that I am interested in "appropriate statistical methods for evaluating computational research studies" rather than "using simulation studies to evaluate statistical methods".

RossGayler, to academicchatter
@RossGayler@aus.social avatar

You can use PREreview to peer review arbitrary preprints on the major preprint archives!

https://prereview.org/about

@prereview @academicchatter @cogsci

easysociology, to Sociology
@easysociology@mastodon.social avatar

Understanding the Marxist Concept of Alienation: A Comprehensive Analysis - Easy Sociology

https://easysociology.com/sociological-perspectives/marxism/understanding-the-marxist-concept-of-alienation-a-comprehensive-analysis/

@sociology
@academicchatter
@academicsunite

RossGayler,
@RossGayler@aus.social avatar

@easysociology @sociology @academicchatter @academicsunite

Does anyone have any opinions about whether these easysociology.com articles are AI generated?

brodriguesco, to random
@brodriguesco@fosstodon.org avatar

Has any of you Linux user had a Linux command segfault if launched via system() or system2() but working fine by itself outside of R?

RossGayler,
@RossGayler@aus.social avatar

@brodriguesco do you run out of swap space in the lead-up to the segfault?

Mehrad, to NixOS
@Mehrad@fosstodon.org avatar

A question for and folks:

I want to start a R project and I want to keep everything about this project static and frozen in time. I know nix-shell is a thing, but:

  1. is there a better/smoother approach?

  2. is it possible to also have a service/daemon running in that nix-shell ? The editor I use (Rstudio) has desktop and server versions, and I would rather have the server version running on a beefy remote machine and I ssh into it.

Boosting is highly appreciated :)

RossGayler,
@RossGayler@aus.social avatar

@Mehrad

@brodriguesco is this a use case for {rix}?

jkanev, to mastodon
@jkanev@fediscience.org avatar

Could we maybe establish a common hashtag for publications? Like or ? So everybody who announces a paper uses this hashtag?

By combining multiple hashtags you could then follow a stream of publications in your science community directly.

What do you think?

RossGayler,
@RossGayler@aus.social avatar

@jkanev I follow OR OR and it's very quiet.

RossGayler,
@RossGayler@aus.social avatar

@Brendanjones @jkanev I agree with all those toots. Being able to follow a more sophisticated search would help because the things I want to follow need the OR of a bunch of terms to capture the different ways people describe it, ANDed with some NOTs to weed outthe majority of stuff I'm not interested in.

I added a like to the github issue.

RossGayler, to random
@RossGayler@aus.social avatar

@brodriguesco I just cast a quick (and very inexpert) eye over your https://github.com/b-rodrigues/rix repo. I see the claim that "With Nix, it is essentially possible to replace {renv} and Docker combined."

What level of dependency can't be addressed by Nix? Obviously not hardware. What about OS version?

MattCrumpLab, to random
@MattCrumpLab@fosstodon.org avatar

Noticed a issue when rendering a citation for a quarto article. I write my name with initials in the author yml:

author: "Matthew J. C. Crump"

The bibtex citation generated at the bottom of the article merges the initials onto the last name.

author = {J.C.Crump, Matthew}

It should be

author = {Crump, Matthew J. C.}

Anyone come across this and know of a solution?

RossGayler,
@RossGayler@aus.social avatar

@MattCrumpLab is that actually a quarto issue? In my writing workflow, RStudio fetches the bibtex citation from the zotero reference manager - so quarto doesn't touch the references other than to render them. If I were having your problem and the name was correctly formatted inside zotero then I would suspect something to do with the zotero export process (converting from internal format to external bites format). Of course, your setup might be completely different.

RossGayler,
@RossGayler@aus.social avatar

@MattCrumpLab Ah. That looks like a design problem. Looking at the YAML example it's easy to see how that could go wrong - there's no information to indicate where words between the first and last name belong. Too bad if your name is (Matthew J.C.) (de Crump). Could you group the names with quotes or something along that line?

RossGayler,
@RossGayler@aus.social avatar

@MattCrumpLab Good luck!

(Is there such a thing as a unicode or html nonbreaking space you could use to append the initals to the first name?)

RossGayler, to mathematics
@RossGayler@aus.social avatar

I want to graphically demonstrate the effect of some algebraic operators in a high-dimensional complex-valued vector space. I will be picking out a small number of example vectors, applying the operators to yield new vectors, and looking at what has happened in terms of the angles between the vectors.

The source vector space may be very high dimensional (say, 1000). Each element of the source vector is constrained to have magnitude 1. That is, each complex value has only one degree of freedom - the phase angle.

I am interested in the angles between vectors in the source space. In these high-dimensional spaces two vectors chosen at random are almost always very close to orthogonal.

I am interested in the angular relations (measured in the high-dimensional source space) between a a small number of vectors and 2 or 3 mutually orthogonal reference vectors.

I want to project the high-dimensional source space onto a real-3-sphere so that the angles of interest are maintained sufficiently well to be visually interpretible. I don't really care what happens to the angle between the other vectors.

I would greatly appreciate any pointers to how I might define and implement such a projection.
(Bonus if you can suggest an R package or code to do this.)

RossGayler,
@RossGayler@aus.social avatar

Thanks @janeadams

having thought about the problem some more since posting the request I suspect that what makes my application unique is the requirement for a wildly nonhomogeneous projection. The projection/embedding techniques i have seen so far treat all points in the input space equally. In my application I want what happens in one local neighbourhood to occupy almost all of the output space while the remainder of the input space is compressed to occupy almost none of the output space.

RossGayler,
@RossGayler@aus.social avatar

@janeadams your query about whether spatial relationships preserve angular relationships is also relevant. In my application, knowing that vectors are pointing in opposite directions is important. Unfortunately, mapping angles to distances in a way that preserves nearness doesn't capture that point because large angles (including direct opposites) get mapped to large distances - so there's nothing to indicate that a sufficiently large distance is actually the opposite direction to some direction of interest.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • tacticalgear
  • rosin
  • Youngstown
  • mdbf
  • ngwrru68w68
  • slotface
  • khanakhh
  • ethstaker
  • everett
  • kavyap
  • thenastyranch
  • DreamBathrooms
  • magazineikmin
  • anitta
  • osvaldo12
  • InstantRegret
  • Durango
  • cisconetworking
  • modclub
  • cubers
  • GTA5RPClips
  • tester
  • normalnudes
  • Leos
  • provamag3
  • megavids
  • lostlight
  • All magazines
    •