Congratulations to Durk Kingma and Max Welling for winning the inaugural ICLR Test of Time Award for their amazing work on Auto-Encoding Variational Bayes, the paper that proposed Variational Autoencoders! arxiv.org/abs/1312.6114

How and why does the highly adaptive lasso break the "curse of dimensionality"?! 1/🧵

Good morning. At some point this summer, perhaps quite soon, AI at Meta will be releasing a LLaMA-3 model with 400B parameters. It will likely be the strongest open-source LLM ever released by a wide margin. This is a thread about how to run it locally. 🧵

In talking to policy makers and AI researchers, I realised there's a fact agreed upon by all researchers, but understood by almost no policy makers. This uncomfortable fact is why AI policy is hard.

linkedin.com/pulse/more-les…

Hello friends and foes are you ready for some effect size standardisation discourse well too bad because you're getting it anyway! (unless you scroll away please don't) Quick thread on a new commentary-like preprint with Protzko @protzko.bsky.social, Daniël Lakens, Malte Elson, & Ruben C. Arslan

It's out in Psychological Review! In this paper, we propose a method for testing whether theories explain empirical phenomena. Thanks Riet van Bork, Adam Finnemann, @JonasHaslbeck, Han, Jill, Jan Sprenger, and @DennyBorsboom for making this possible! tinyurl.com/52h6e42f

You know what’s nice about meta-analysis studies? They’re perfect. There is just nothing wrong that can go there and every single one is pure, informative, impact. In this new preprint, we valiantly sought to falsify this universally held belief. 1/7

New paper and software alert! arxiv.org/abs/2408.14620 Interested in modern mediation analysis methods with machine learning and multivariate mediators? Take a look at this joint work with Richard Liu, @nickWillyamz , and Kara Rudolph Short 🧵...

In other words while non-parametric models might make the asymptotic limit more robust it also pushes the asymptotic limit further away from practical relevance!

Happy to announce some new work with my student Kaitlyn Lee! arxiv.org/abs/2501.04871

If your learning algorithm is based on correlation rather than causation, it will struggle with overfitting. To understand something is to identify its minimal sufficient causal mechanisms. Parsimony isn't just elegance, it's generalization robustness.

Curious to hear others' views on this. Also finding similar results in my own work... 🫣 Alicia Curth it reminds me a bit of your "Doing Great at Estimating CATE?" work, but with a different angle.

My colleagues and I are looking for engineers with a mixture of MLOps, applied ML, and software engineering experience to work on some new med-tech. Please reach out for more information, if interested.

Drowning in junk science: Is there any hope at all? statmodeling.stat.columbia.edu/2025/04/17/dro…

We won! Grateful to have joined my new colleagues for GNVC!

x.com/i/article/1946…