A Focus on Foundations: Halıcıoğlu Data Science Institute Hosts NSF-Funded AI Institute

Article Content

The National Science Foundation (NSF) recently announced that the Halıcıoğlu Data Science Institute (HDSI) at UC San Diego is the future home of The Institute for Learning-enabled Optimization at Scale, or TILOS.  The $20M Artificial Intelligence (AI) hub will foster research and focus on “making impossible optimizations possible” at both scale and in practice.

TILOS research will pioneer learning-enabled optimizations transforming chip design, robotics, networks and other application use domains that are vital to the nation’s health, prosperity, and welfare.  NSF funding includes a five-year, $220 million investment to create 11 multidisciplinary, cross-institutional research institutes to serve as hubs for higher education and federal agencies.  In addition, non-profit and industry partners, including Intel, will work together to advance AI-focused research and workforce development initiatives.

A major part of TILOS is foundational research in AI and optimization.  "Optimization is both a science and a technology,” says Rajesh K. Gupta, Founding Director of HDSI.  “This new AI institute will not only discover new science at the interface of AI and optimization, but also deliver it to real-world practitioners as a technology: measuring it to improve it, with benchmarking and a roadmap of progress.”

Faculty and researchers alike will identify applications of basic research corresponding to three distinct application areas: chip design, robotics, and networks.  As an institute, TILOS emphasizes working partnerships between the Foundations Team and the applications experts in these three application areas. Each of the TILOS application areas produces rich, complex datasets that challenge research on foundations to stretch the boundaries of AI-enabled optimization, including areas where improvements could make a dramatic impact on everyday life. 

The TILOS Foundations Team aims to discover key principles and new methods in the interplay of AI and optimization, and the TILOS team is comprised of leading researches with expertise in optimization, machine learning, and algorithms.  HDSI researchers have taken a leading role in shaping the TILOS institute. Professor Arya Mazumdar is a co-PI and co-lead of the Foundations Team. Professor Yusu Wang is Associate Director of Research for TILOS. HDSI Distinguished Scientist Michael Pazzani will serve as the Managing Director of the Institute. 

According to Professor Mazumdar, “Foundation research in TILOS plays two major roles.  First, unveiling the most fundamental mathematical questions related to optimization and machine learning; these range from demystifying the success of deep neural networks to finding limits of distributed optimization methods.  Second, interfacing with the applications team and providing algorithmic solutions to the most pressing and practical problems with tools from AI and optimization.”  He concludes, “Our team at HDSI comprises leading experts specializing in machine learning and decision making, including Professors Mikhail Belkin and Yian Ma.”

Mathematical research of AI and optimization in TILOS is further structured into sub-teams that work on problems of different natures.  All are foundational, but requiring different sets of mathematical tools.

Optimization problems usually come in two flavors. In one case, the parameters of the problem are discrete, where the optimal object needs to be found from a finite but large set.  Take for example, the famous traveling salesman problem. In another case, the parameters of the problems are continuous; for example, finding the best temperature and pressure required for a chemical experiment. The techniques used for these two flavors of optimization problems are vastly different. One sub-team of Foundation researchers is bridging the gap between these two paradigms by using methods designed for one in another. Yet, another sub-team is looking at techniques of distributed computation to process large-scale big-data optimization problems of both discrete and continuous flavors.

HDSI researchers are involved in all sub-teams of Foundation research.  Foundation Team members will bring several modern views to elucidate the interplay of AI and optimization, such as the relation between Deep Learning and non-convex optimization, optimization on manifolds or other non-Euclidean spaces, federated learning, and so on.  These topics are timely, made urgent by the rapid advancements in AI and the prevalence of related data.

HDSI Professor Tara Javidi is co-leading the Networks research team. “Our demand for data and reliable connectivity is exponentially growing, and sustainable scalability of our modern information infrastructure is at stake,” Javidi said. “Optimizing the design and operation of networks will not only enable new capabilities in autonomous driving, augmented and virtual reality for telemedicine, and robotics, but will also result in tremendous energy savings to help combat climate change.”

HDSI Professor Yusu Wang, an expert on topological and geometric data analysis at UC San Diego, will serve as the TILOS Associate Director for Research leading efforts that bring together foundational and applied research.  “Optimization problems are ubiquitous, affecting so many sectors in society.  HDSI plays a key role in TILOS by contributing broad expertise from data science and machine learning and connecting this expertise to both theory of optimization as well as applied domains.”

UC San Diego’s researchers will also partner with other faculty and researchers from several other institutions, including the Massachusetts Institute of Technology (MIT), National University, University of Pennsylvania (UPenn), University of Texas, and Yale University.  HDSI will also collaborate with other UC San Diego faculty and researchers in Computer Science and Engineering, and the Jacobs School of Engineering. The Foundation Team, in particular, is collaborating closely with researchers from MIT, Yale, and UPenn on the topics of parallelizing optimization methods, fairness in AI, and new optimization algorithms for complex systems with very large number of parameters,  To learn more about the TILOS project and the team, visit the TILOS AI website.

The UC San Diego HDSI faculty research team includes: Professor Mikhail Belkin; Professor in Electrical and Computer Engineering, Jacobs Family Scholar, and HDSI Faculty Fellow Tara Javidi; Assistant Professor Yian Ma; Associate Professor and TILOS co-PI Arya Mazumdar; Distinguished Scientist Michael Pazzani; and, Professor Yusu Wang.

The Halıcıoğlu Data Science Institute at UC San Diego advances the field of data science by exploring principles, methods and tools that enable us to understand the nature of digital data and the interactions of this growing field to existing disciplines of human inquiry.  As the home of Data Science academic programs at UC San Diego, HDSI works collaboratively across multiple academic fields and with industry partners to explore the scientific foundations and real-world applications of data science to address and solve society’s most pressing problems.