Skip to main content
Event

PRIMA — Invited Presentation on Neurosymbolic Learning for Science

Dr. Omar Costilla Reyes was invited to present his team’s work on neurosymbolic learning for science to European Union colleagues on June 28, 2022. The talk was part of the PRIMA research project.

Title slide of the “Understanding the World Through Code” team’s work on neurosymbolic learning for science, shared ahead of the June 28, 2022 PRIMA presentation. Credit: Image courtesy of Dr. Omar Costilla Reyes.

Invited presentation for PRIMA

PRIMA, the Partnership for Research and Innovation in the Mediterranean Area, is a European Union research program whose main objective, as Dr. Omar Costilla Reyes quoted it, is to devise new research and innovation approaches to improve water availability and sustainable agricultural production in a region heavily stressed by climate change, urbanization, and population growth. At the invitation of Rubén Vera-Rodríguez, Dr. Costilla Reyes was scheduled on June 28, 2022 to present his team’s work on neurosymbolic learning for science to colleagues at the European Union as part of the project, and he announced it the day before.

What neurosymbolic learning for science means

Neural networks learn patterns directly from data but are difficult for a scientist to read, while symbolic programs are explicit, readable sets of rules and operations that can state what is already known. Neurosymbolic learning combines the two: learned components sit inside a program structure that carries scientific domain knowledge, such as known constraints and high-level concepts, so that results stay interpretable to researchers. The slide shown here is the title slide of the team’s work, “Understanding the World Through Code,” a multi-institution project led by Armando Solar-Lezama at MIT with colleagues at Rice, UT Austin, Caltech, the University of Pennsylvania and Stanford, in which Dr. Costilla Reyes was a research scientist. The position paper “Neurosymbolic programming for science,” posted to arXiv in October 2022 and presented at the NeurIPS 2022 AI for Science workshop, set out the approach, and he helped organize a neurosymbolic programming summer school at Caltech in July 2022.

A long collaboration

Vera-Rodríguez, of the Universidad Autónoma de Madrid, had been a collaborator since Dr. Costilla Reyes’s doctoral research, co-authoring the 2018 footstep-recognition paper in IEEE Transactions on Pattern Analysis and Machine Intelligence, and Dr. Costilla Reyes thanked him for his expertise in the security themes of that work when he completed his PhD.

Topics