←
Frank Hutter618 posts
Frank Hutter avatar

Frank Hutter

@FrankRHutter

Founder of @Prior_Labs, Prof of #machinelearning at ELLIS Institute Tübingen & Uni Freiburg, 3x ERC Grant holder, EurAI & ELLIS fellow. All opinions are my own.

◷ Joined July 2018▧ 118 Photos and videos
Frank Hutter avatar
Frank Hutter@FrankRHutter·Sep 15

The data science revolution continues. TabPFN-3.5 is live, claiming SOTA beyond the vanilla IID small data setting and tackling data complexities that data scientists face everyday: ✅  grouped data ✅  temporal data ✅  strings in tables ✅  strong uncertainty calibration ✅  one model, whether it's classification or regression, 100 or 1.000.000 rows ✅  scaling to 20k features (benchmarked thoroughly to 6k) ✅  up to 6x faster inference with TabPFN-3.5-fast, plus FP-8 support on API Model report: storage.googleapis.com/prior… Try it today! 👉 Open-source repo: github.com/PriorLabs/tabpfn 👉 API: ux.priorlabs.ai/

Post attachmentPost attachment
0:39
Frank Hutter avatar
Frank Hutter@FrankRHutter·Nov 11

I'm super excited about this course! It's been an honour to record with @AndrewYNg 🎉 Thanks to everyone involved in this, and especially to @tuanacelik and Eliott Kalfon for teaching this with me!

Make predictions on tabular data without training a new model for each dataset. In Tabular Foundation Models for Data Science Agents, built with @prior_labs, you'll give an AI agent a tabular foundation model as a prediction tool. Enroll for free: bit.ly/4sMtNOz

2:10
Frank Hutter avatar
Frank Hutter@FrankRHutter·Sep 24

It was great to be on @MLStreetTalk, speaking about tabular foundation models with Tim Scarfe. 2 hours of great questions about TabPFN! Full episode: lnkd.in/efWfXC72

> "Tabular data is very dirty. I think that's one of the reasons deep learning took so long to do well for it." We spoke with @FrankRHutter of @prior_labs about TabPFN, a foundation model that makes predictions on tables in one forward pass, trained entirely on synthetic data!

4:46
Frank Hutter avatar
Frank Hutter@FrankRHutter·Aug 18

Super excited about our first work on relational learning, continuing to push on open source!

We’re happy to announce our first release in relational learning at Prior Labs, continuing our commitment to open science. We open-source three pieces of software that we expect to accelerate research in the field towards meaningful, real-world impact. First and foremost, we release 𝗥𝗲𝗹𝗔𝗿𝗲𝗻𝗮-α: a unified framework for running and comparing baselines on RelBench v1 tasks. Based on learnings from tabular benchmarks like TabArena, we are standardizing data loading, evaluation protocols, tuning regimes, and adding support for systems with custom tuning. We also open-source 𝗧𝗮𝗯𝗣𝗙𝗡-𝗥𝗲𝗹: our relational harness for TabPFN-3. We initialize the (living) RelArena-α leaderboard with TabPFN-Rel and a comprehensive set of baselines. The rankings at the time of release are: • 𝗧𝗮𝗯𝗣𝗙𝗡-𝗥𝗲𝗹 is the No. 1 model submission • 𝗥𝗧-𝗣𝗹𝘂𝗥𝗲𝗹 is the No. 1 system submission Last but not least, we open-source an alpha version of the 𝗥𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝘃𝗲 𝗜𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲 (𝗥𝗣𝗜): enabling you to easily specify prediction tasks on your own relational database and then run any RelArena-α model, like TabPFN-Rel, in a few lines of code, all bundled as a simple PyPI package. • Read the full model report: arxiv.org/abs/2608.16319 • GitHub repository (give us a ⭐️): github.com/PriorLabs/relaren… • Announcement: priorlabs.ai/blog-posts/intr… • Docs: docs.priorlabs.ai/capabiliti… Thanks to the contributions from: @adrihayler, @KNfloege, @AlanArazi1536, @_rishabhranjan_, @jure, @LennartPurucker, @FrankRHutter & @noahholl

Post attachment
Frank Hutter avatar
Frank Hutter@FrankRHutter·Jul 17

3/4 Planning horizon: We can now boldly tackle long-term high-risk high-gain research moonshots. As an 18-month old startup you cannot tackle problems that require years of research; as a frontier AI lab, now we can.

Frank Hutter avatar
Frank Hutter@FrankRHutter·Jul 17

2/4 Scale: @SAP is the perfect partner, with an enormous scale of tabular data and distribution. This will allow us to build dramatically better models and directly deploy them at scale. We’ll also substantially scale up the team; this is the time to join!

Frank Hutter avatar
Frank Hutter@FrankRHutter·Jul 17

Today it is official: our acquisition by SAP has closed. Prior Labs continues as an independent frontier AI lab - same team, same offices, same mission, open models, backed by €1B+. What changes? The scale and the planning horizon. 🧵 1/4

Post attachment
RAISE Summit avatar
⟳ Frank Hutter retweeted
RAISE Summit@RaiseSummit·Jun 29

Deep learning transformed how machines understand text and images. But the data that actually runs most enterprises, the rows and columns inside spreadsheets and databases, has largely missed that revolution. For decades, structured data has powered decisions across healthcare, finance, and business, yet it's been analyzed using many of the same approaches long after AI advanced everywhere else. That is beginning to change. A new generation of Tabular Foundation Models is bringing frontier AI to structured data, delivering faster, more accurate predictions without the task-specific training traditional models require. It's already powering real-world applications, from early disease detection to financial risk modelling, and redefining what's possible with enterprise data. This July 8, @FrankRHutter, Founder & CEO at @Prior_Labs, takes the Grace Hopper Stage for his keynote, "The Foundation Model Revolution for Structured Data." Drawing on the latest advances behind Prior Labs' TabPFN, he'll explore where the technology stands today, why this represents a fundamental shift for enterprise AI, and what's next for the future of structured data. If foundation models transformed text and images, what happens when they finally transform the data your business actually runs on? Final ticket price increase on July 2. 🔗 Secure your ticket:  hubs.li/Q04mY1fL0

Post attachment
Frank Hutter avatar
Frank Hutter@FrankRHutter·May 12

Available three ways: 👉 Open weights for research & internal evaluation 👉 API with TabPFN-3-Plus & Thinking Mode 👉 Enterprise licensing (incl. AWS SageMaker, Azure AI Foundry, on-prem) Model report in replies. 🧵 5/5 With ❤️ from Prior Labs

Frank Hutter avatar
Frank Hutter@FrankRHutter·May 12

Speed at scale: 👉Reduced KV cache (~7GB per estimator at 1M rows) 👉Cached predict: 1–3 orders of magnitude faster than cold fit+predict 👉30ms prediction latency on H100 up to 100k rows 🧵 4/5

Post attachment
Frank Hutter avatar
Frank Hutter@FrankRHutter·May 12

TabPFN-3 Thinking Mode (API) takes this further: 200+ Elo over the next-best non-TabPFN method on TabArena, 420 Elo on the largest-data subset. 99% win rate against tuned & ensembled XGBoost 🧵 3/5

Post attachment
Frank Hutter avatar
Frank Hutter@FrankRHutter·May 12

A few highlights: 👉 Pareto-dominant on TabArena 👉 SOTA up to 1M training rows 👉 Up to 1000x faster cached inference 👉 SHAP via shapiq up to 120x faster 👉 SOTA on many-class, RelBenchV1, and text-tabular 🧵 2/5

Post attachment