The NSF AI Planning Institute brought domain experts in astronomy, high energy physics, biophysics, climate science, and engineering together with experts in AI, statistics, and computer science. The common tools used by all members and visitors served to unify these very disparate topics and move forward research in all areas, so that progress in one area could immediately translate to another without wheel reinvention.

One example of this is super-resolution simulations, developed to simulate small scale structure in the universe. These
simulations are physics-based models of the Universe and its contents.
Even with enormous computational resources there are tradeoffs between
volume and resolution. We developed deep learning methods which
seamlessly extend the dynamic range of these calculations. Generative
neural networks trained on high resolution (HR) models are
conditionally applied to low resolution models, leading to full super
resolution (SR) simulations. We showed how statistical properties at
SR are remarkably consistent with full HR, but can be orders of
magnitude larger and more complex than standard techniques allow.
Deep learning can therefore combine and extend modelling of physics
processes, offering a route to simulation of the Universe beyond
current models of superclusters and galaxies and into the realm of
stars and planets. This work was the subject of press releases which
highlighted the work of the AI Planning Institute, and was picked up
by over fifty different news outlets. We learned that this technique is used not only by our cosmologists, but also by climate scientists (who call the method "downsampling"), engineers, and biologists. This led to one of 5 well-attended conferences that the Institute hosted.

An interdisciplinary seminar series was formally established, with
speakers covering domain areas that included climate forecasting,
extragalactic astronomy, nuclear fusion power, molecular biology,
hardware security, quantum chemistry, and particle physics. The
unifying themes were those of the Institute, the links between AI and
Physics. The seminars included 40 minutes
of open discussion. Participants included undergraduate and graduate
students, faculty members from five different CMU departments and also
from sixteen other institutions, including Historically Black Colleges
and Universities as well as national labs, and companies
such as Qualcomm and Amazon. Seminars were weekly
throughout the whole academic year and there were 102 unique
participants, with an average attendance per seminar of 31.
The seminar series was linked with a course on AI in Physics, which
provided opportunities for training for the diverse participants.

The Institute hosted a summer REU program starting in 2021. Seven students were accepted, and six of them were from under-repsented groups. Each was embedded in a research group and produced posters at the end of their 7 weeks of research. The Institute housed them and provided them with 5-6 social events so that the program provided students with a well-rounded experience that offered not just access to state-of-art AI research but also the opportunity to interact with peers and become comfortable in a research setting. The success of this program led us to repeat it for all the years of the Planning Institute, and it has now been combined with other summer programs at CMU to become an independent NSF supported REU.

Among the disseminations of results were interactive
demonstrations and presentations of AI techniques in astrohysics to
classes of students at Arsenal Middle School. The school body is
composed of 85% underrepresented minorities and 98% of students
qualify for free/discounted lunches. Lively discussions occurred
during the presentations on the role of AI in modern life and how the
generative models used in Minecraft are different from cosmological
simulations. Similar interactions occurred during a high school teacher training program that we set up that is now in its 3rd year of operation.

We developed a course for undergraduates that was taken by students at CMU as well as other schools such as South Carolina State University and University of Maryland, Baltimore County. 

Last Modified: 05/03/2024
Modified by: Scott Dodelson