Wibisono Wins NSF CAREER Award to Develop New Framework for Algorithm Design

For his innovative work on reimagining how algorithms are designed, Andre Wibisono has won a Faculty Early Career Development (CAREER) Award from the National Science Foundation (NSF).
Wibisono, assistant professor of computer science, will use the $657,704, five-year grant to develop a groundbreaking framework that bridges the gap between continuous mathematical systems and practical computer algorithms. The NSF CAREER award is a prestigious honor for young faculty members and supports the early career activities of teachers and scholars who are considered most likely to become the academic leaders of the future.
Modern machine learning systems, which power everything from artificial intelligence to scientific simulations, depend on solving complex computational tasks efficiently. While many powerful algorithms are inspired by continuous-time mathematical systems, converting these mathematical models into practical, step-by-step computer algorithms has typically required intensive case-by-case analysis. Wibisono's research aims to develop a systematic framework for this translation process.
"Think of it like converting a smooth motion into a series of snapshots," Wibisono said. "Computers can only work with discrete steps, like frames in a movie, but many of our most powerful mathematical ideas are based on continuous motion. Converting between these two worlds while preserving the mathematical properties we care about is a crucial challenge in modern computing and one we aim to solve."
Wibisono’s innovative approach combines tools from multiple mathematical fields, including dynamical systems, geometry, probability and information theory.
“Mathematics provides the language to precisely describe the phenomena that we want to capture in the algorithms, and these mathematical fields provide the tools that we can use to study the algorithms,” Wibisono said. “For example, probability and information theory have a rich history and a wealth of results that are very useful to develop our algorithmic results. Sometimes in our study of modern algorithms for machine learning, we not only borrow results from these fields, but also imbue them with new geometric or optimization meaning.”
The research has broad applications beyond traditional computing, extending to areas such as algorithmic game theory, modern generative AI systems such as Stable Diffusion and OpenAI’s DALL-E, and problems such as quantifying uncertainty, robustness, fairness, and privacy in machine learning systems.
"The dynamical perspective of algorithm design helps researchers understand how various algorithms are related and which ones they should use for their problems," Wibisono said. "Instead of simply applying an algorithm as a black box and hoping it works, researchers can better understand and apply the principles behind these algorithms."
This framework could revolutionize how we develop algorithms across many fields, providing a more intuitive and powerful approach to solving complex computational challenges in science and technology.
The project includes significant educational components, including a new curriculum on modern algorithms for machine learning. Wibisono will also lead a collaborative effort to create a comprehensive catalog of algorithms and dynamics for optimization and sampling, which will serve as a valuable resource for students, practitioners, and researchers.
"My goal is to demystify the mathematics behind modern machine learning algorithms, which are often steeped in the abstract language of stochastic processes," Wibisono said. "The class aims to highlight and make as transparent as possible the simple principles that are operating underneath these algorithms, so that the students can understand how and why the algorithms work."