Lightnews — Scholar-powered news

Jan van Gemert

@jvgemert.bsky.social

1.2K followers 210 following 93 posts

Head of the Computer Vision lab; TU Delft. - Fundamental empirical Deep Learning research - Visual inductive priors for data efficiency Web: https://jvgemert.github.io/

jvgemert.github.io

Posts Media Videos Starter Packs

Jan van Gemert @jvgemert.bsky.social · 3d

I'm also interested! I've been thinking about Marr in times of deep learning :)

Jan van Gemert @jvgemert.bsky.social · 5d

The Monodepth Monologues? ;)

Reposted by Jan van Gemert

David Picard @davidpicard.bsky.social · 5d

Now you can either do 2 things: Adding more coils, bigger magnets, an iron core, and engineering all that until you get a useful product that changes the world.

Or you can try to understand what is going on, like Michael Faraday did and as later James Clerk Maxwell formalized.

I want the latter.

2 2 23

Jan van Gemert @jvgemert.bsky.social · 5d

Well, constraints can encourage creativity. Eg, haiku; sonnets, dogma movies, social-media character limits 😀.

Jan van Gemert @jvgemert.bsky.social · 5d

Do scientific papers need to be boring? 🤔

Science News @scinews.bsky.social · Dec 15

7. How to write consistently boring scientific literature.

“Hell – is sitting on a hot stone reading your own scientific publications”
Erik Ursin, fish biologist

Thanks @fattebertj.bsky.social

Open Access
nsojournals.onlinelibrary.wiley.com/doi/epdf/10....

1 5

Jan van Gemert @jvgemert.bsky.social · 6d

Max Planck's quote about science progress.. ;)

Reposted by Jan van Gemert

Melanie Mitchell @melaniemitchell.bsky.social · 7d

In 2016 Hinton predicted that AI would replace all radiologists in five years. Ten years later, why hasn't it happened? This post is a great explainer.

www.understandingai.org/p/ai-isnt-re...

AI isn't replacing radiologists

Radiology combines digital images, clear benchmarks, and repeatable tasks. But demand for human radiologists is at an all-time high.

www.understandingai.org

11 150 350

Jan van Gemert @jvgemert.bsky.social · 8d

Happy birthday 😁

1 1

Jan van Gemert @jvgemert.bsky.social · 9d

Sounds interesting! Do you perhaps have a freely accessible version? 🤔

Jan van Gemert @jvgemert.bsky.social · 16d

Sorry.. slow person here 😅.. why would in-plane rotation be expected? And why particularly for DINO? 🤔

Jan van Gemert @jvgemert.bsky.social · 19d

Let's go! Can you send me an email :)

Reposted by Jan van Gemert

Georg Bökman @bokmangeorg.bsky.social · 20d

A simple argument for equivariance at scale: 1) At scale, token-wise linear layers dominate compute. 2) Token-wise linear equivariant layers implemented in the Fourier domain are block-diagonal and hence fast.

3 1 8

Jan van Gemert @jvgemert.bsky.social · 20d

Very nice work! Thank you for posting! 🙂

1 2

Reposted by Jan van Gemert

David Picard @davidpicard.bsky.social · 23d

😱😱😱 @jvgemert.bsky.social are you ok? You could have told us all you were a ghost from the beginning!👻

2 2 12

Jan van Gemert @jvgemert.bsky.social · 23d

Hahaha, 🤣.
Ok, the game is up. Busted. Ghost-busted.

Jan van Gemert @jvgemert.bsky.social · 28d

I could perhaps think of a simpler hypothesis, but Occam just ran out of blades 😕

en.m.wikipedia.org/wiki/Razor_a...

Razor and blades model - Wikipedia

en.m.wikipedia.org

Jan van Gemert @jvgemert.bsky.social · 28d

Oops, you did it again..

Jan van Gemert @jvgemert.bsky.social · Sep 7

Also the mask-rcnn argument isn't very strong: DETR doesn't need proposals.

Jan van Gemert @jvgemert.bsky.social · Sep 7

I'm confused about the claim here that (Self-attention?) in the transformer binds the features as claimed here 🤔.

I have the hypothesis that it doesn't (and, thus also disagree with the hypothesis of Scholte/De Haan 🙂 )

Jan van Gemert @jvgemert.bsky.social · Aug 2

Great video! I very much disagree with the field of ontologies, but it's very well done 🙂

Jan van Gemert @jvgemert.bsky.social · Aug 1

There are diminishing returns. Stop trying to improve review quality by excessive extra work.

Remember the irreducible Bayes Error: If the task has inherent subjectivity, it will not help to get better quality measurements.

David Picard @davidpicard.bsky.social · Aug 1

Everybody misses the 1-page rebuttal.

These lengthy forum style comments are a nightmare: a nightmare for the authors who spend way too much time writing them, a nightmare for the reviewers who spend too much time understanding them, a nightmare for the ACs who will have to summarize all. Stop it!

Rémi Flamary @rflamary.bsky.social · Aug 1

As a #NeurIPS AC I did a little stat. Paper reviews+responses in my batch contain 515k characters and 75k words. This is without including papers themselves or authors/reviewers discussions. So please be nice to your AC, we are doing our best. Also I miss the one page PDF response of old ICML.

1 3

Reposted by Jan van Gemert

Itai Yanai @itaiyanai.bsky.social · Jul 29

🚀 Today we're launching the 'Night Science Institute', a non-for-profit organization to lead a cultural shift in science! You may say we're dreamers 😉, but we think we're not the only ones. Perhaps today you will join us to make the Day Science and Night Science parts of the process live as one!

1 45 120

Jan van Gemert @jvgemert.bsky.social · Jul 25

Interesting observation.. Goodhart's law strikes again?

Although, for me, the more senior I've gotten, the less I care about just another "top venue" publication, and the more i value the scientific quality 🙂

Jan van Gemert @jvgemert.bsky.social · Jul 23

Guilty 😅

Oded Rechavi @odedrechavi.bsky.social · Jul 22

When the professor edits the student's draft

Jan van Gemert @jvgemert.bsky.social · Jul 20

Reminds me somehow of Plankton..