For good probability predictions, you should use post-hoc calibration. With Eugène Berta, Michael Jordan, and Francis Bach we argue that early stopping and tuning should account for this! Using the loss after post-hoc calibration often avoids premature stopping. 🧵1/

I am pleased to share at lnkd.in/eQjxvSvM slides, podcast and an essay on my lecture: “Ten Simple Rules for Mathematical Writing” Since its original delivery in a slide presentation at MIT (2002), it has been referenced widely and used in mathematical writing courses.

Hi all, if you're attending ICML (Vancouver) or UAI (Rio de Janeiro), I'm happy to share some news from the lab! Please check it out -- and feel free to drop by or shoot me a line if any of it sounds intriguing. 1/5 "Counterfactual Graphical Models: Constraints and Inference"

5/5 Last but definitely not least, I’m honored to be giving a keynote on Wednesday (7/23) titled "Towards Causal Artificial Intelligence." For details, see: auai.org/uai2025/keynot… Here’s a short abstract: While many AI scientists and engineers believe we are on the verge of

Amazing rise of an underdog journal , using a new streamlined model of review and publication. Econometric Journal’s Impact factor is now the same as Econometrica’s. Please consider submitting your work there!

Neuschwanstein Castle, Germany 🇩🇪

📘 My book Impact Evaluation in Firms and Organizations is officially out today with The MIT Press @mitpress.bsky.social! An accessible, non-technical introduction to impact evaluation (& causal machine learning) designed for practitioners & students, with use cases in R & Python: mitpress.mit.edu/9780262552929/…

In economics, elite concentration, intellectual gatekeeping, and disciplinary homogeneity is higher than in all other fields. This has been well-documented by now. Here are three papers on this: Fourcade, Ollion and Algan (2015), “The Superiority of Economists”: Economics is a

World Models are all the hype, but it is difficult to get started? Check this easy to use codebase from alfred nguyen, Franz Srambical (not at neurips cuz no capacity) and Mihir Mahajan

Podcast Series for the Causal-ML book: youtube.com/playlist?list=….

In a recent work (causalai.net/r136.pdf), we examined whether LLMs are potential sources of probabilistic knowledge (rung 1 of Pearl's hierarchy), which led to the benchmark at llm-observatory.org. The answer was no, which was surprising and poses fundamental challenges for

The first seminar in the Rational Intelligence Seminar Series (RISS) by Sebastian Zezulka is now available on YouTube. Check it out and don't forget to subscribe to our YouTube channel for future seminars. More info: ri-lab.org/riss/ youtu.be/YhCsBKmbpM0?si…

Your reposting this old note is very timely in view of Imbens's recent review article: annualreviews.org/content/journa…, which contains a causal diagram in almost every page. As you know, the pledge to avoid diagrams has been a required oath of allegiance in the potential outcome camp,

xLSTM excels in time series forecasting: arxiv.org/abs/2509.01187 . Introduces "stochastic xLSTM" (StoxLSTM). "StoxLSTM consistently outperforms state-of-the-art baselines with better robustness and stronger generalization ability." TiRex shows that xLSTM is time series king.

Having worked on statistical foundations of #LLMs for 2 years, excited to share that 2 papers got accepted to top stats journals: Algorithmic Bias in RLHF (arxiv.org/abs/2405.16455) → JASA Robust detection of watermarks (arxiv.org/abs/2411.13868) → JRSSB

SHAP is the Swiss Army Knife of ML interpretability: If you would only bring one tool, SHAP is a versatile choice. But like the Swiss Knife, it has limitations. Learn more about SHAP and its limitations with the book "Interpreting Machine Learning Models with SHAP".

🇩🇪 Next up: Bavaria Germany’s largest state, where alpine peaks, baroque cities, and centuries of tradition meet. Let’s explore 🧵

I am looking for PhD students and postdocs to join our group in beautiful Munich at the Technical University of Munich and Helmholtz Munich. PhD application info: drive.google.com/file/d/1PM-y0s… postdoc application info: drive.google.com/file/d/1ripNnM… Please help me spread the word! 🤗

Sensitivity Analysis for Treatment Effects in Difference-in-Differences Models using Riesz Representation. arxiv.org/abs/2510.09064

Sensitivity Analysis for Causal ML: A Use Case at Booking.com. arxiv.org/abs/2510.09109