Show me your argparser's size, and I'll see your code's vibe

Berlin Mathematics Research Cluster MATHplus organizes the "TES Conference on Mathematical Optimization for Machine Learning" at Zuse Institute Berlin (Sep 13-15), preceded by a summer school (Sep 11-13).

Just got my copy of "Little Book of Deep Learning" by François Fleuret. Can't wait to dive into this concise 150-page treasure! 📖

I have written a series of posts about the idea of Interactive Classification (from Interactive Proof Systems) for designing interpretable classifiers: stephanw.net/blog/2023/FI_s… In this setup, we prove lower bounds on the information content of the selected features.

This is the 1st rigorous treatment (and 3rd verification) I've seen IMO - this is great for AI safety! It means that LLMs are doing *exactly* what they're trained to do — estimate next-word probability based on data. Missing data? P(word)==0 So where is the AI logic? 1/🧵

Excited to be at #NeurIPS2023! We'll be presenting our work on Accelerated Riemannian Min-Max Optimization at the OPT+ML workshop during poster session 2 from 3:00 - 4:00 pm. arxiv.org/abs/2305.16186

🌟 Join our Team in Berlin 🌟 We are seeking highly motivated PhD students to work on (efficient) Deep Learning, preferably with strong math/CS background and PyTorch experience. Happy to answer questions here, via DM or at #icml2024! Apply at iol.zib.de/openings! Please RT

In our paper "Convergence and Trade-Offs in Riemannian Gradient Descent and Riemannian Proximal Point" at #ICML2024, we examine a blind spot in the Riemannian opt literature: Most works simply 𝘢𝘴𝘴𝘶𝘮𝘦 that the iterates stay in a bounded set. This is a problem because 1/8

Strong picture!

DeCAF won the #ICML Test of Time Award 2024! Big congrats to trevordarrell (my PhD advisor at MIT), and Jeff Donahue. 🎉 You may not heard of DeCAF, but it is everywhere! DeCAF stands for Deep Convolutional Activation Features. Published ten years ago, the DeCAF paper is a

Schufa Holding GmbH has a paper on fairness #icml2024, lol. Next paper will be on transparency, i bet?

In a new AI Snake Oil essay by me and Sayash Kapoor, we do a deep dive into AI existential risk probability estimates. We find that these forecasts are just feelings dressed up as numbers, and even the best-run, well funded, time intensive forecasting efforts result in a range of

Join our poster session *today* at #ICML2024 in the TF2M Workshop. Looking forward to the inspiring discussions.

arXiv -> alphaXiv Students at Stanford have built alphaXiv, an open discussion forum for arXiv papers. alphaXiv You can post questions and comments directly on top of any arXiv paper by changing arXiv to alphaXiv in any URL!

Many criticize LLMs like GPT-4o for failing tasks like counting letters in “strawberry.” This isn’t a failure but a design choice, as models aim to be helpful and interpret intent. GPT-4o might assume the query is about “berry” vs. “bery,” not counting ‘r’s. Here are 2 examples.

.OpenAI is hiring ML engineers for a new multi-agent research team! We view multi-agent as a path to even better AI reasoning. Prior multi-agent experience isn't needed. If you'd like to research this area with Kevin Stone and me fill out this form: jobs.ashbyhq.com/openai/form/oa…

Doing good science is 90% finding a science buddy to constantly talk to about the project.

Short summary of our recent paper on adversarial defenses vs watermarking together with Sai Ganesh Nagarajan Berkant Turan and Grzegorz Gluch - you gotta get at most one. pokutta.com/blog/research/… #machinelearning #watermarks #adversarialdefense

Poster: Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation arxiv.org/abs/2501.19328 Blogpost: maxzimmer.org/blog/2025/capt… w/ Jan Pauls, Berkant Turan, S. Saatchi, Philippe.ciais, Sebastian Pokutta & F. Gieseke