| | Dear colleagues. It's been an exciting month for the CISE research community. On behalf of everyone in CISE, I want to congratulate Andrew Barto from the University of Massachusetts Amherst and Richard Sutton from the University of Alberta for winning the 2024 Association for Computing Machinery A.M. Turing Award. Barto and Sutton are recognized as pioneers in reinforcement learning (RL), a key part of modern artificial intelligence and decision-making systems. Barto's work, significantly supported by the U.S. National Science Foundation, laid the foundation for RL and continues to impact areas like chatbots and supply chain optimization. As I learned of the award, it reminded me of how much of the work we see having an impact in today's world started as ideas that, at the time, were exploring concepts that required imagination and risk-taking. Indeed, one of the challenges we face as a community and as a funding agency is how to choose among the immediate opportunities in front of us — opportunities that present a tangible impact and opportunities whose ideas spark the imagination but have no specific result in sight. The fact is, of course, that we need to do both in appropriate measure. As I look through the CISE portfolio, I can easily find examples, like RL or deep learning, cryptography, wireless communications, etc., that started in the lab and took decades to mature. But, there are also many examples in today's NSF research portfolio like the NSF-led National Artificial Intelligence Research Institutes or the NSF Security, Privacy, and Trust in Cyberspace program or many other programs where the impacts are far more immediate, whether through an iconic research result or by enabling transition to practice, or because a student moved from the research lab to a startup or a company where their academic work found fertile ground. My request to you is to help us identify and develop these stories. CISE is turning 40 years old next year (2026). As part of our 40 years of progress, we would like to surface the gestation, evolution and real-world impact of key ideas in computing. Stay tuned for more information. In the meantime, keep exploring, inventing and creating the future! Greg Hager
Assistant Director for CISE | | News and updates  Supercomputers Work Together for Big Science Simulations. University of Wisconsin–Madison researchers used several NSF-funded supercomputers to simulate plasma turbulence in space. The research was made possible through NSF-funded ACCESS allocations, providing powerful computing resources for simulations. | | | Events  Spring 2025 Advisory Committee for Cyberinfrastructure (ACCI). The ACCI meeting will provide a platform for experts to advise the NSF on strategies to advance state-of-the-art cyberinfrastructure that enables significant advances across all fields of science and engineering. May 20–21, 2025. 10 a.m. – 4 p.m. | | | Community Spotlight  Rajeev Alur, the recipient of the 2024 Donald E. Knuth Prize, is the Zisman Family Professor of Computer and Information Science and the founding director of the ASSET Center for Trustworthy AI at the University of Pennsylvania. Alur specializes in developing methods and tools that help system designers specify how a system should behave and ensure it meets those expectations. His work has improved the safety and accuracy of technologies like network protocols, multiagent systems, and embedded control software. As AI plays an increasingly critical role in modern computing, Alur is dedicated to ensuring its trustworthiness for society. Alur and his collaborators have been exploring how logical rules can improve reinforcement learning (RL), a method used to train autonomous systems like robotic arms and self-driving cars. Traditional RL relies on reward functions to guide learning, but designing these rewards for complex tasks is difficult and prone to errors. The success of RL can also be highly sensitive to how these rewards are structured. To address this, Alur's research focuses on specification-guided RL, which uses logical languages, like linear temporal logic, to define objectives instead of relying on reward functions. This approach makes it easier to specify complex tasks clearly and provides precise mathematical interpretations. Their research shows that when no time-bound is specified for completing a task, it is mathematically impossible to guarantee success when using RL with logical specifications. However, when such a time-bound is fixed, they have developed efficient learning methods that take advantage of logical structures, improving performance and reducing the amount of training data needed. | | |
Division of Information and Intelligent Systems (IIS)
Supports research and education on the interrelated roles of people, computers, and information to advance knowledge of artificial intelligence, data management, assistive technologies, and human-centered computing. Office of Advanced Cyberinfrastructure (OAC)
Supports the conceptualization, design, implementation, and operation of research cyberinfrastructure to advance and transform research and education in science and engineering. | | | | |
No comments:
Post a Comment