Profile avatar
aleksei-tiulpin.bsky.social
#AI, #MedicalImaging, #Uncertainty, and now #FoundationModels. Leading @imedslab.bsky.social at the University of Oulu. Sports, nutrition, and health enthusiast. Amateur musician and fly-fisherman in my free time.
14 posts 39 followers 92 following
Regular Contributor

Aithyra is hiring! We are looking for multiple starting PIs in life sciences and AI/ML. We offer highly competitive packages and salaries among the highest in Europe. www.oeaw.ac.at/aithyra/news...

πŸ”₯πŸŽ‰New library: boosting for survival analysis, including multiclass (competing risks) Survival = missing outcomes because limited observation window (common in medicine, marketting...) soda-inria.github.io/hazardous Gives very fast boosted-trees for survival

If you think MS dragon copilot will solve medicine, first reflect on how great MS Teams is.

1/ Introducing ACE (Amortized Conditioning Engine)! Our new AISTATS 2025 paper presents a transformer framework that unifies tasks from image completion to BayesOpt & simulator-based inference under *one* probabilistic conditioning approach. It's Bayes all the way down!

My new paper "Deep Learning is Not So Mysterious or Different": arxiv.org/abs/2503.02113. Generalization behaviours in deep learning can be intuitively understood through a notion of soft inductive biases, and formally characterized with countable hypothesis bounds! 1/12

Science is a creative endeavor, an art, aiming at discovery of new knowledge, ways to think, and ways to create. When you are in a rat race, you might get distracted by the race itself and the β€œwhat” while forgetting the β€œwhy”. Without it, there is no art anymore. It's just papers.

My coauthors and I WANT our textbook to be used for LLM training, as we want our way of thinking to be promoted by AI

πŸ‘‰ You are probably still reporting your results wrong. πŸ‘‰ Take a look at the fresh release of my #opensource library for model comaprison: imedslab.github.io/stambo/index.... πŸ’‘ And by the way, do not forget to subscribe to Random Samples: randomsamples.substack.com.

πŸš€ Proud to launch my substack: Random Samples. I will be writing about #Stats, #DataScience, and #MachineLearning. No BS & no hype: technical and conceptual topics broken down to be understood. Sign up here: randomsamples.substack.com.

David Lynch has passed away. R.I.P.

The circle of fifths is so underrated when people try to learn music.

The beginning of the year seems to be quite intense. I am working on a new university-level course on statistics. Already recorded some of the videos for students, and planning to experiment with AI-based narration in the future.

While everyone was / is at #NeurIPS2024, I've been going through some "heavy times". 8 String guitars are really something different. It feels like I am starting over (I play guitar for 19 years).

It feels like statistics is obfuscated by a lot of unnecessary mathematical abstractions. I feel like a proper modern course on stats should include: Python, Pandas, Bootstrap, and a Permutation Test. Working on one right now.

Found slides by Ankur Moitra (presented at a TCS For All event) on "How to do theoretical research." Full of great advice! My favourite: "Find the easiest problem you can't solve. The more embarrassing, the better!" Slides: drive.google.com/file/d/15VaT... TCS For all: sigact.org/tcsforall/

I am catching up late, but it seems that high-fidelity generation is all you need.

I am sol glad to see #AI research impacting many other fields. It has never been so impactful, and it is likely that in the future we will see even bigger impacts of it.

The UI here is quite glitchy, but I guess it will be fixed.