Topic

AI for Science

How machine learning and program synthesis help scientists discover, test, and explain ideas, with models people can read.

1 episode

About AI for Science on Interpretable

This topic is about using machine learning and program synthesis to help scientists discover, test, and explain ideas. Its episode is Episode 000, in which the host, Omar Costilla Reyes, explains why the show is called Interpretable. The talk opened a two-day tutorial on AI for code and science in the Neurosymbolic Programming for Science series.

The idea is neurosymbolic programming: models that combine neural networks with symbolic programs written in a domain-specific language. The episode explains what a program is, how program synthesis differs from training a neural network, and why writing a model as code makes it readable, since much of scientific knowledge already takes the form of code: equations, rules, and procedures. The paper behind the talk, Neurosymbolic Programming for Science, from the NeurIPS 2022 AI for Science workshop, argues that these techniques could accelerate scientific discovery, that scientists can build their own domain knowledge into the models, and that the outputs are interpretable. It uses behavior analysis as a worked example and discusses opportunities and challenges across the natural and social sciences. The episode and the paper are listed below.

Episode

References

  1. Jennifer J. Sun, Megan Tjandrasuwita, Atharva Sehgal, Armando Solar-Lezama, Swarat Chaudhuri, Yisong Yue, Omar Costilla-Reyes (2022). Neurosymbolic Programming for Science(opens in a new tab). NeurIPS 2022 Workshop on AI for Science. Discussed in Why Interpretable? Code as the Language of Science.