Ten new interdisciplinary research projects have won funding from Princeton University’s Schmidt DataX Fund, with the goal of spreading and deepening the use of artificial intelligence and machine learning across campus in order to accelerate discovery.
The 10 faculty projects, supported through a major gift from Schmidt Futures, involve 19 researchers and several departments and programs, from computer science to politics.
The projects explore a variety of subjects, including an analysis of how money and politics interact, discovering and developing new materials exhibiting quantum properties, and advancing natural language processing through the automatic construction of novel knowledge bases.
“We are excited by the wide range of projects that are being funded, which shows the importance and impact of data science across disciplines,” said Peter Ramadge, director of the Center for Statistics and Machine Learning (CSML). “These projects are using artificial intelligence and machine learning in multi-faceted ways: to unearth hidden connections or patterns, model complex systems that are difficult to predict, and develop new modes of analysis and processing.”
CSML is overseeing a range of efforts made possible by the Schmidt DataX Fund to extend the reach of data science across campus. These efforts include the hiring of data scientists and overseeing the awarding of DataX grants. This is the second round of DataX seed funding with the first in 2019.
The winning projects and research faculty are the following:
Discovering developmental algorithms
Bernard Chazelle, Eugene Higgins Professor of Computer Science
Eszter Posfai, Assistant Professor of Molecular Biology
Stanislav Y. Shvartsman, Professor of Chemical and Biological Engineering and the Lewis Sigler Institute for Integrative Genomics
“Natural algorithms” is a term used to described dynamic, biological processes built over time via evolution. This project seeks to explore and understand through data analysis one type of natural algorithm, the process of transforming a fertilized egg into a multicellular organism.
MagNet: Transforming Power Magnetics Design with Machine Learning
Tools and SPICE Simulations
Minjie Chen, Assistant Professor of Electrical Engineering and the Andlinger Center for Energy and the Environment
Niraj Jha, Professor of Electrical and Computer Engineering
Yuxin Chen, Assistant Professor of Electrical and Computer Engineering
Magnetic components are typically the largest and least efficient components in power electronics. To address these issues, this project proposes the development of an open-source machine-learning based magnetics design platform in order to transform the modeling and design of power magnetics.
Multi-modal Knowledge Base Construction for Commonsense Reasoning
Jia Deng, Assistant Professor of Computer Science
Danqi Chen, Assistant Professor of Computer Science
To advance natural language processing, researchers have been developing large scale, text-based commonsense knowledge bases, which help programs understand facts about the world, but these data sets are laborious to build and have issues with spatial relationships between objects. This project seeks to address these two limitations by using information from videos along with text in order to automatically build commonsense knowledge bases.
Generalized clustering algorithms to map the types of COVID-19 response
Jason Fleischer, Professor of Electrical and Computer Engineering
Clustering algorithms are made to group objects but fall short when the objects have multiple labels, the groups require detailed statistics, or the data sets grow or change. This project addresses these shortcomings by developing networks that make clustering algorithms more agile and sophisticated. Improved performance on medical data, especially patient response to COVID-19, will be demonstrated.
New Framework for Data in Semiconductor Device Modeling, Characterization, and Optimization Suitable for Machine Learning Tools
Claire Gmachl, Eugene Higgins Professor of Electrical Engineering
This project is focused on developing a new machine-learning driven framework to model, characterize and optimize semiconductor devices.
Individual Political Contributions
Matias Iaryczower, Professor of Politics
This project proposes to use micro level data on the individual characteristics of potential political contributors, characteristics and choices of political candidates, and political contributions made in order to answer questions on the interplay of money and politics.
Building a Browser-Based Data Science Platform
Jonathan Mayer, Assistant Professor of Computer Science and Public Affairs, Princeton School of Public and International Affairs
Many research problems at the intersection of technology and public policy involve personalized content, social media activity and other individualized online experiences. This project, which is a collaboration with Mozilla, is building a browser-based data science platform that will enable researchers to study how users interact with online services. The initial study on the platform will analyze how users are exposed to, consume, share, and act on political and COVID-19 information and misinformation.
Adaptive Depth Neural Networks and “Physics” Hidden Layers: Applications to Multiphase Flows
Michael Mueller, Associate Professor of Mechanical and Aerospace Engineering
Sankaran Sundaresan, Norman John Sollenberger Professor in Engineering, Professor of Chemical and Biological Engineering
This project proposes to develop data-based models for complex multi-physics fluids flows using neural networks in which physics constraints are explicitly enforced.
Seeking to Greatly Accelerate the Achievement of Quantum Many-Body Optimal Control Utilizing Artificial Neural Networks
Herschel Rabitz, Charles Phelps Smyth '16 *17 Professor of Chemistry
Tak-San Ho, Research Chemist in the Department of Chemistry
This project seeks to harness artificial neural networks to design, model, understand and control quantum dynamics phenomena between different particles, i.e. atoms and molecules.
(Note: This project also received DataX grants in the previous round of funding.)
Discovery and Design of the Next Generation of Topological Materials Using Machine Learning
Leslie Schoop, Assistant Professor of Chemistry
Bogdan Bernevig, Professor of Physics
Nicolas Regnault, Visiting Research Scholar in the Department of Physics
This project aims to use machine learning techniques to uncover and develop “topological matter,” a type of matter that exhibits quantum properties, whose future applications can impact energy efficiency and the rise of super quantum computers. Current topological matter’s applications are severely limited because their desired properties only appear at extremely low temperatures or high magnetic fields.