Hosted by Computing Sciences at Berkeley Lab, the 2026 Deep Learning for Science (DL4SCI) Summer School is a five-day intensive program bringing together researchers and engineers to explore the latest advances in deep learning and generative AI (GenAI), with a special emphasis this year on foundation models, reasoning, and agentic AI for scientific discovery.

The school will feature in-depth lectures, research talks, and hands-on tutorials centered on emerging approaches to foundation models for science. Topics will span the end-to-end lifecycle—data, training at scale, adaptation, and evaluation—along with sessions on reasoning-centric workflows and agentic systems. Theory will be covered, but the program emphasizes practical application, including best practices for running deep learning on high-performance computing systems. Local and international AI experts will lead interactive labs and provide modern code and content. Attendees will gain the tools to design and deploy robust model- and agent-based approaches for scientific problems.

The Summer School will also facilitate networking and collaboration through breakout sessions, group activities, and optional poster sessions. These forums will allow participants to engage directly with instructors and peers, fostering vibrant discussions on how current research trends—particularly in foundation models, reasoning, and agents—can be leveraged in scientific domains. By the end of the program, attendees will be equipped with the tools and expertise necessary to implement, evaluate, and scale modern AI solutions in their research.