AI institute for advances in optimization

The National Science Foundation Division of Computer and Network Systems awarded a $1 million Cooperative Agreement to Georgia TECH Research Corp on October 1, 2026, under the Computer and Information Science and Engineering program (CFDA 47.070). The AI Institute for Advances in Optimization (AI4OPT) fuses artificial intelligence and optimization to address challenges in energy systems, supply chains, and semiconductor manufacturing specified in America's AI Action Plan. The institute will develop optimization proxies that deliver near-optimal solutions to complex problems in milliseconds and learning-to-optimize techniques to reduce problem dimensionality, enabling substantial computational speedups. AI4OPT will stabilize and optimize grid resources to support energy demand growth, provide end-to-end planning and control capabilities for supply chains and semiconductor manufacturing processes, and deliver K-12 education programs including data science and AI summer camps, interactive MOOCs, and video game design camps focused on energy and supply chain systems. The institute will also train educators, policy makers, and parents, and expand industrial collaborations through innovation labs, a venture studio, and platform development to accelerate technology transfer. Work is performed in Atlanta, Georgia with a period of performance from October 1, 2026, through September 30, 2031. Georgia TECH Research Corp serves as the research arm of Georgia Institute of Technology.

Abstract

The AI institute for advances in optimization (AI4OPT) fuses AI and optimization to achieve outcomes that cannot be delivered by the two fields independently. AI4OPT directly addresses challenges from america's AI action plan in energy systems, supply chains, and semiconductor manufacturing. The science and technology pioneered by AI4OPT will deliver key enablers to stabilize the grid and optimize existing and future grid resources, as needed to support the growth in energy demand. These advances in science and technology are also essential for the end-to-end planning and control of supply chains and manufacturing processes, including for the semiconductor industry which faces daunting challenges. In education, AI4OPT focuses on k-12 programs, directly in line with the executive order on advancing artificial intelligence education for american youth. The institute will make data science and AI summer camps available nationwide through interactive moocs, offer camps to design video games for energy systems and supply chains, and train educators, parents, and policy makers. On translation to practice, AI4OPT will expand its numerous industrial collaborations and technology transfers into an end-to-end innovation pipeline, featuring innovation labs, a venture studio, and platform development to maximize impact. To achieve these societal goal, AI4OPT leverages and builds on top of (1) the concept of optimization proxies, which provide feasible near-optimal solutions to complex optimization problems in milliseconds, including certificates of solution quality and (2) learning to optimize techniques that leverage AI to reduce the dimensionality of optimization problems by orders of magnitude, enabling substantial speedups. AI4OPT aims at providing a step change in the fusion of AI and optimization to address the fundamental societal challenges mentioned earlier, as the world continued to evolve at a rapid pace. IT will develop the concept of AI for control-aware optimization, recognizing the need to be aware of underlying control systems and optimize over systems of systems to make high-quality decisions. Optimization proxies will be generalized to this broader setting and expanded to deep decision rules for multi-stage stochastic optimization, the learning of optimization proxies in sobolev space, diffusion-based optimization proxies, and foundational models for ML-guided optimization. AI4OPT also recognizes that optimization techniques have much to offer to AI in their fusion. IT will explore more effective embedding of AI models in optimization models, the robust training of AI models using mixed-integer optimization, quantization-aware training methods, and resource-efficient supervised learning through improved batching and exploiting similarity. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are planned for this award.