Event: Horizons in Biosciences & Informatics Seminar Series (HBISS)
What: The Foundation Model Revolution for Structured Data
When: Tuesday, September 22nd, 8a PT; 11a ET; 5p CEST; 8:30p IST
Where: Blue Marble Space Zoom
Who: Dr. Frank Hutter; Co-CEO of Prior Labs, PI at ELLIS Institute Tübingen, Director of the ELLIS unit at University of Freiburg
Summary:
Tabular data, the rows and columns of spreadsheets and databases, sits at the core of most questions in science, business, and medicine. Yet while deep learning transformed text and images, tabular data remained the domain of gradient-boosted trees for two decades. This talk introduces tabular foundation models, a new class of AI built for structured data. TabPFN, published in Nature in 2025 and validated across hundreds of independent studies, delivers reliable predictions in seconds without task-specific training, outperforms previous methods by a wide margin, and keeps improving at the exponential pace typical of foundation models. It already powers thousands of real-world applications, from early disease detection to financial risk modeling, and the latest version,TabPFN-3, extends state-of-the-art performance to adjacent tasks such as time series forecasting and relational data. Real-world impact, however, requires more than accuracy. I therefore also discuss the trustworthiness of tabular foundation models, including interpretability, robustness, fairness and causal reasoning that moves beyond correlation to predicting the effects of interventions. Finally, I show how tabular foundation models plug into agentic workflows to allow state-of-the-art machine learning without writing a singleline of code.
Biography:
Frank Hutter is a founder and co-CEO of Prior Labs, the company behind Tabular Foundation Models. He is also a PI at the ELLIS Institute Tübingen and Director of the ELLIS unit Freiburg, where he is a Full Professor of Machine Learning (currently on leave), and a three-time recipient of grants from the European Research Council (ERC). He holds a PhD from the University of British Columbia and is a Fellow of EurAI and ELLIS. Before founding Prior Labs, Frank pioneered the field of automated machine learning (AutoML), co-authoring the first AutoML book, the first AutoML MOOC, and the widely-used AutoML systems Auto-WEKA and Auto-sklearn, as well as organizing 20+ workshops on AutoML-related topics and co-founding the AutoML conference. In recent years his focus is on tabular foundation models.
About HBISS:
The purpose of HBISS is to foster interdisciplinary conversation, education, and collaboration. Experts in biosciences and informatics are invited to speak once a month and engage with members of the Open Science Data Repository’s Analysis Working Group (OSDR-AWG) and the public to facilitate discussion of cutting-edge research, techniques, and methodology.
From March 2022 through February 2025, the HBISS series was designed exclusively for an internal NASA audience. Beginning in April 2025, the series was opened to the public and the broader OSDR-AWG community. To stay in the loop, join the AWG.
Questions, or want to be removed from this HBISS listing?
Contact HBISS organizer ryan.t.scott@nasa.gov