About the Workshop

The Schmidt AI for Science program at the University of Toronto will host a three-day workshop in Toronto exploring how AI foundation models can solve interdisciplinary scientific problems. A foundation model is a general-purpose AI/ML model that has already learned information suitable for diverse tasks; state-of-the-art results show that a foundation model trained with data from one scientific domain (e.g., astronomy) may be suitable for problems in another scientific domain (e.g., chemistry). This workshop combines lectures with hands-on tutorials and culminates in a hackathon.

In the hackathon, attendees will form teams that test and utilize foundation models for science; teams will compete for prizes in categories such as innovative solutions, best team spirit, and most robust model. All local expenses, including four nights' accommodation, transportation, and meals, will be covered by the organizers. Attendees will be responsible for funding their travel to and from Toronto, along with any extra days outside of the workshop days.

Participants are expected to attend the full three-day program (approximately 9:00 AM to 5:00 PM daily) and actively contribute to the hackathon. Participants will also attend social events organized outside of workshop hours.

Goals for the workshop

  • Understand the state-of-the-art foundation models across various disciplines.
  • Gain hands-on experience implementing foundation models across different scientific domains.
  • Develop interdisciplinary problem-solving and communication skills for tackling complex scientific challenges.
  • Form cross-institutional research networks that can continue beyond the workshop.

Target Audience

We are seeking approximately 50 participants with diverse scientific backgrounds who are interested in learning and applying cross-domain AI solutions. Researchers who plan to or currently use foundation models to unlock new insights in scientific research are particularly encouraged to apply. Proficiency in Python and at least one deep learning framework (PyTorch, TensorFlow, or JAX) is expected, though various levels of familiarity with foundation models are welcome, as tutorials will accommodate different skill levels.

First-year fellows interested in foundation models are especially encouraged to attend. Ideal candidates will possess technical programming abilities, domain-specific knowledge and an interest in working on scientific use cases that may not directly relate to their current research.