Why Interpretable? Code as the Language of Science
Dr. Omar Costilla Reyes
NeurIPS · 2022Paper discussed
Why scientific knowledge looks like code, and why that makes AI readable.
Speaker
Host of Interpretable, founder of equ healthcare
Dr. Omar Costilla Reyes is the host of Interpretable and the founder of equ healthcare. A researcher in neurosymbolic AI and computational neuroscience, he is the corresponding author of "Neurosymbolic Programming for Science" (NeurIPS 2022 AI for Science workshop), written with collaborators at MIT CSAIL, Caltech, and UT Austin.
Omar Costilla Reyes is the host of Interpretable. His website describes him as the founder and CEO of equ healthcare, whose work is research and engineering for explainable AI in health and scientific discovery, and says he did his PhD at the University of Manchester. The research areas it lists include explainable machine learning, digital phenotyping, symbolic regression, computational neuroscience, neurosymbolic AI, and neurosymbolic program synthesis.
His episode is built around the paper Neurosymbolic Programming for Science, a NeurIPS 2022 AI for Science workshop paper with Jennifer J. Sun, Megan Tjandrasuwita, Atharva Sehgal, Armando Solar-Lezama, Swarat Chaudhuri, and Yisong Yue, also available on arXiv. The paper argues that neurosymbolic programming techniques could accelerate scientific discovery. These models combine neural and symbolic components, can incorporate a scientist's domain knowledge and experimental context, and can produce interpretable outputs. The authors identify opportunities and challenges in fitting current models to scientific workflows, use behavior analysis as a worked example, and aim to make the methods useful across the natural and social sciences.
Episode 000, Why Interpretable? Code as the Language of Science, explains the idea behind the show's name. A model that is partly a program can be read, tested, and corrected, because much of scientific knowledge already takes the form of code: equations, rules, and procedures. The talk opened a two-day tutorial on AI for code and science in the Neurosymbolic Programming for Science series, and it covers program synthesis, how it differs from training a neural network, and how neurosymbolic programming links machine learning with scientific knowledge. The episode is filed under Interpretable AI and AI for Science.
Dr. Omar Costilla Reyes
NeurIPS · 2022Paper discussed
Why scientific knowledge looks like code, and why that makes AI readable.