From the 23rd to 26th of March, the third annual Conference on Parsimony and Learning (CPAL) brought together a global community of researchers to explore the power of parsimony, revealing the simple, low-dimensional structures that underlie machine learning, signal processing, optimization, and beyond. After previous editions at Stanford and Hong Kong, this year’s event was hosted in Tübingen, by the ELLIS Institute Tübingen, in conjunction with the Tübingen AI Center and Max-Planck-Institute for Intelligent Systems. In this year’s edition, talks and tutorials moved beyond theoretical foundations to tackle the pressing challenges of scaling, safety, and efficiency in the era of Large Language Models.
Over four days, the event progressed from tutorials on foundational methods and practical techniques to high-level keynotes on feature learning, uncertainty quantification, and institutional machine learning. Renowned speakers from leading international institutions explored how low-dimensional structures can make massive models more efficient and robust.
CPAL 2026 featured Rising Stars awards, highlight talks, two poster sessions, and great opportunities for networking.
This event was sponsored by the HKU Institute for Data Science, CISPA Helmholtz Center for Information Security, OPTML, and Michigan State University.
Find out more about the full program and details here.
(c) Roberto Montebello