@trappmartin.bsky.social
Assistant Prof in Prob ML @ KTH 🇸🇪 WASP Fellow & ELLIS Member Ex: Aalto Uni 🇫🇮, TU Graz 🇦🇹, originally 🇩🇪. — https://trappmartin.github.io/ — Reliable ML | UQ | Bayesian DL | tractability & PCs
The deadline to submit papers to ProbML's proceedings and workshop tracks is **March 20**!
📣Check out the CfP and submit here: probml.cc/call/!
Only 10 days left to submit your work to ProbML (formerly AABI)!
👉 probml.cc/call/
To conclude my political rant. The US and Russia are not examples of a new world order; if anything, they are pushing for the old world order that we had centuries ago, when my grandmother was sheltering to avoid getting killed by US bombardments.
This is not a world we should want to get back to.
So, instead of ruling by power, as Russia and the US do, Europe moved towards a system in which every country has less power.
It would be completely ridiculous to think of a setting in the EU where a country wants to annex another, like the US with Canada and Russia with Ukraine.
Why is the European Union as a concept relevant? After pretty much a destruction of Europe, the model of the EU established peace in a region troubled by wars. Independent countries & democracies coexist peacefully and collaborate by accepting loosing power to an independent political body.
One where countries with military power (Russia/US) aim to control the world order by … power. As a European, this is clearly a step backwards from where we used to be, and a recipe for uncertainty and wars.
If anything, we should move towards something like the European Union, a true new model.
Like many Europeans, I am troubled by the recent wars in Ukraine and the Middle East. And I am shocked by the narratives pushed by the US.
From what I read, the US is pushing for a “new” world order, which really should be called the old world order.
The Guardian on our new study, which shows that global heating is significantly gathering speed. Our efforts to overcome our fossil fuel addiction should do the same.
www.theguardian.com/environment/...
Fascinating, thanks.
Is there a “behind the scenes” blog post/document?
Guter Bericht der Tagesschau zu unserer neuen Studie, gemeinsam mit dem US-Statistiker Grant Foster.
Die Messdaten belegen: die Erderwärmung hat sich sogar beschleunigt. Jede Verzögerung beim Ausstieg aus fossiler Energienutzung halte ich für unverantwortlich.
www.tagesschau.de/wissen/klima...
After those excellent suggestions, I wonder why not write a script to crawl all my arxiv submissions, update their reference lists with every “relevant” paper, push to arxiv and do this automatically every month.
Sparked by a humorous comment here, I wonder how people self-cite?
I have seen people citing almost every paper they publish, people asking to be cited because of group affiliation, and people who self-cite fairly little, incl. myself. What’s your approach and why?
If you are interested working with me as a postdoc on probabilistic machine learning, please consider reaching out and applying for the postdoc call by digital futures (link below).
🔗 www.digitalfutures.kth.se/call/up-to-t...
"He is from [MASK] [MASK]" → "San York"? dLLMs fail because they ignore token dependencies. This Factorization Barrier arises from a structural misspecification: models are restricted to fully factorized outputs. We break this barrier with CoDD, enabling coherent parallel generation. 🚀
Are you a grad student interested in computational #astrophysics? Check out our upcoming #FlatironCCA Pre-Doctoral Program — applications are due March 13, 2026. https://www.simonsfoundation.org/2023-flatiron-institute-center-for-computational-astrophysics-pre-doctoral-programhow-to-apply/ #science
If you missed the @auai.org deadline but have some awesome work you wanna share with the world, #ProbML submissions deadline is just around the corner! And it’s in Seoul 🇰🇷, meaning good food is guaranteed.
In light of the current funding situation (worldwide), a modest proposal: instead of pouring billions of dollars into GenAI claiming "it *could* accelerate science and research," consider putting 1% of that amount in what *will* accelerate science and research. Namely, funding science and research.
Same here. In the past, I used Vim almost exclusively, but now that I don’t do this anymore, I am out of training and not fast enough for larger code bases.
We are looking for Postdoctoral Fellows to join the Helsinki region IT research community.The deadline for applications is March 22nd, 2026 at 11:59 PM (23:59 UTC+02:00) More info in the 🔗
www.hiit.fi/hiit-postdoc... #EarlyCareerResearcher #AcademicSky #postdocposition
I would definitely agree and have been telling folks this for a long time, but stopped as the Python fan base can be somewhat difficult at times.
There seems to be some reasonably active library on this out there: github.com/THU-BPM/Mark..., which again I have not tested.
Watermarking LLM output is of course not bulletproof, but seems to gain some adoption. So I assume it works to some extent. (I am not an expert on this)
There are several approaches for this (see watermarking LLMs), some of which are actively used by major ML conferences, notably ICML is one of them. I don’t know which ones are the best and would need to look into it myself too.
The #ECCV2026 paper registration deadline is THIS WEEK!
NO registration by the deadline → NO submission. NO exceptions.
Deadline details are available on the webpage.
Deadlines: eccv.ecva.net/Conferences/...
Two days to StanCon 202 oral presentation abstract submission deadline!!!
Probably, but it also makes it much easier to identify those that are “cheating” (assuming LLM use is not allowed) by watermarking assignments.
A list of robot benchmarks and datasets and some notes on them! A quick blog post to give some color to my last few posts on how robot policy evaluation is done
Benchmarking robotics is an interesting topic with still no good answers. open.substack.com/pub/itcanthi...
How to make me feel old, without telling me that I am old.
… 4 weeks later… [exhausted voice] “see, it’s easy”
@cpaxton.bsky.social might know.