Jan van Gemert

I really enjoy finding gaps in our understanding of text-book deep learning. The RNN Homunculus: Long-range input vanishes RNN gradients. Gates (GRU, LSTM) help, but gates are themselves an RNN. Ie: gates are not sufficient. We need short-range data to learn gates first. arxiv.org/abs/2610.14279

In addition to the research findings, this is the type of research I really enjoyed doing 😊 "On Hyperparameter Tuning on the Test Set" arxiv.org/abs/2610.05902

I only know the basics about flow matching/diffusion, but this is really nice! Moore, Escher, Penrose: A Conformal Golden Braid arxiv.org/abs/2610.02210

I'm not sure I agree with @giorgio.gilest.ro's logic that this ends up with technofeudalism. Yes LLM-written papers are flooding journals, but that's because journals are and always were a bad tool for filtering for insight. Our response should be to try something different.

I'm burnt out of discussing "AI". I don't care for the press releases from big companies. I hate having to comment immediately. I dislike "hot takes". I'm worried we, as scientists, are losing touch with scientific values. We've become customers, salivating at ads marketed to investors. 1/15

Open Position for an Assistant/Associate Professor in Visual Analytics: careers.tudelft.nl/job/Delft-As.... Apply if you are excited about interactive exploration, analysis, and interpretation of complex data. Please also forward this position to anyone in your network who is interested.

I never wanted to share my slides before the lecture because it ruins my Socratic teaching style 🍵 But.. I found out they are circling in the interwebs, so, I might as well share them myself then 🙂. Then at least they are up to date 😊. jvgemert.github.io/teach.html

Our fifth workshop on Meta-science for Machine Learning is on Sept 22. metascienceforml.github.io Registration info in the link.

A Delfie at the European Conference on Computer Vision #ECCV2026 :)

@eccv.bsky.social #ECCV2026 How can I organize which posters to see where? The app cuts off paper titles (ie: unreadable and unusable). The website's schedule is great, but has no poster IDs next to the titles. Will we get a handy hardcopy program, with titles, times, and ids?

Computer vision people, where are you??? 👀

#ECCV2026 Workshop on Empirical Theory: tracing back the causal relationships in our scientific papers. Untroubling the "troubling trends in machine learning scholarship" empiricaltheory.github.io @eccv.bsky.social

📢 We're hiring: Tenure-Track Professor for Philosophy of Machine Learning for Science @unituebingen.bsky.social! We're looking for an outstanding researcher with a strong profile in #philosophy and strong connections to #ML and #AI. Deadline: 30 Sept 2026; more info: uni-tuebingen.de/en/128980 1/3

@vele.bsky.social @cscv-bot.bsky.social I love your arXiv bot ♥️. But.. reading it is difficult due to the author names coming first, so the most important thing is difficult to find: the paper title 🙂. If it was up to me, I would remove the author list completely 😁.

NeurIPS reviewers not liking understanding-based ML work? Consider submitting to our workshop: empiricaltheory.github.io - R1: Reject. Not state of the art. - R2: Reject. Just small toy datasets. - R3: Reject. After reading the paper it's trivial.

The E.T.👽 workshop submission is finally open! We aim at promoting scientific theory building in deep representation learning 2 tracks: 1p storyline & regular 14p full paper Deadline: Aug 1st 🖥️ empiricaltheory.github.io 📜 openreview.net/group?id=the... Let's do science! @eccv.bsky.social #ECCV

Understand-based research vs BOLD numbers. Submit your work to our ECCV workshop on Empirical Theory: empiricaltheory.github.io @davidpicard.eurosky.social @a-lucic.bsky.social m.loog, h.hung, h.pinson.

The #ECCV2026 workshop page with links is now up! eccv.ecva.net/Conferences/...

Honered to get a selfie with our Netherlands Conference on Computer Vision (NCCV) keynote speaker! The brilliant @vickykalogeiton.bsky.social Two zealots of the Data efficiency church in the Religion of Scaling AI 🙏

This piece brilliantly explains why it's not only unnecessary to take seriously the idea that LLMs are conscious, in some cases it is ethically irresponsible.

NEW PAPER. Why do larger networks train better? "Because they contain more candidate *sub*networks that can learn the task" → lottery tickets This popular explanation uses an appealing but misleading metaphor🧵 We propose an intuitive alternative grounded in theory: escape dimensions

I'm taking this opportunity to make some advertising for an ECCV Workshop on empirical theory proposed by @jvgemert.bsky.social empiricaltheory.github.io Yes it's related. Think harder. You can connect the dots, I have faith in you 😉

"Science is not simply a sequence of tasks that can be optimized....The scientific system thrives on inefficiency: redundant efforts, failed attempts, and divergent paths. These are not costs to be eliminated but sources of discovery." Excellent essay! ⬇️ www.science.org/doi/10.1126/...

Did you know that arXiv has an official merch shop? I didn’t, but I think it’s great. 😅 www.bonfire.com/store/arxiv-...

We are hiring! We have exciting PhD opportunity to improve trustworthiness of causal inference at TU Delft, in collaboration with @gelovennan.bsky.social , in the Safe Causal Inference consortium. Apply now! #PhD #CausalInference #MachineLearning #Statistics careers.tudelft.nl/job/Delft-Ph...

Progress in academia

We're looking for a new colleague at @amlab.bsky.social: Assistant Professor in AI for Science 🔬🤖 World-class ML research, Amsterdam's thriving AI ecosystem (ELLIS, startups, big tech), and some of the best academic labor conditions in Europe ❤️ Deadline: May 30 👉 werkenbij.uva.nl/en/vacancies...

- The bitter lesson: Richard Sutton's view on AI - The butter lesson: more butter makes better food @gracekind.net - The batter lesson: crêpes are better than pancakes - The better lesson: AI can potentially improve teaching - The botter lesson: genAI takes over social media

In this incredibly detailed and gracious review of “The Irrational Decision,” @himself.bsky.social situates the book in the web of intellectual history and the war between the AI enthusiasts and skeptics.

Writing forces your brain to coordinate memory, reasoning, and meaning-making simultaneously. Every time you write, you rewire toward clearer thinking. Every time you let an LLM do it, you rewire toward consumption.

🚨 Happy to announce CVPR@Paris'26 which will take place on June 1st in Paris. The goal of the event is to share a little bit of the conference before it happens. We will have poster sessions as well as several plenary talks by world-class speakers. info: cvprinparis.github.io/CVPR2026InPa...

/s

Great post about how and why do science (and winning a best paper award is not the goal).

A day in the life of… youtu.be/T4Upf_B9RLQ?...

"In the interest of time I should agree with you. But that would just not be me." -every faculty meeting I've been to

Bibliography cleanup is known to be harder than AGI. Same with successfully connecting a laptop with a projector. It's what we humans will be doing long after the robot takeover. 😇

With some trepidation, I'm putting this out into the world: gershmanlab.com/textbook.html It's a textbook called Computational Foundations of Cognitive Neuroscience, which I wrote for my class. My hope is that this will be a living document, continuously improved as I get feedback.

I'd like to propose the following norm for peer review of papers. If a paper shows clear signs of LLM-generated errors that were not detected by the author, the paper should be immediately rejected. My reasoning: 1/ #ResearchIntegrity

Inspiring, though provoking, disagreeable 😉. Great to have @lakens.bsky.social as a keynote at our third metascience for Machine Learning workshop metascienceforml.github.io

Interesting take on the role of publication venues.

Riffing on Ed Leamer’s “sinning in the basement,” I ponder a theoretical reformation in machine learning. (There are ImageNet scatterplots, IYKYK.)

Interesting initiative; started by frustrated scientist; for frustrated scientists.

Meet @dimadamen.bsky.social, Prof of Computer Vision at the @bristoluni.bsky.social 🏴󠁧󠁢󠁥󠁮󠁧󠁿 & ELLIS Fellow working in Multimodal Learning Systems. She aims to inspire more women to lead in #AI by highlighting more role models. Discover her career milestones and goals for the field. #WomenInELLIS

This ambitious line of work is a great example of empirical theory in deep learning! I think we need more empirical theory in ML/DL (in contrast what ML people call 'theory', which is math, but not empirical 🙂). Metascience for ml 🙂 metascienceforml.github.io

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.