CAREER: Automated Machine Learning Compilation for Emerging Models and Platforms

As artificial intelligence becomes ubiquitous in our daily lives, there is a critical need to cost-effectively deploy AI models to diverse environments, such as data centers, smartphones, and web browsers. This project aims to bring a new AI/ML compiler toolchain to help accelerate and automate AI deployments. The project’s novelties are: i) a unified way to represent optimizations in AI models, ii) bringing AI agents together with domain expertise to optimize deployment, and iii) end-to-end solutions for automated model deployment. The project's impacts are three-fold. First, it makes AI/ML models more accessible in a broad set of environments, from the cloud to local devices. Second, it helps improve the overall AI/ML tools and toolchain ecosystem by providing automated solutions to optimize the toolchains that power applications like ChatGPT, Claude, and Gemini. Finally, it helps reduce the effort required to establish AI/ML toolchains for the latest hardware through AI-based automation. This project addresses the challenges that span multiple levels of the ML stack—from adjusting modeling choices and high-level execution strategies—to implementing low-level optimizations across diverse hardware architectures. The investigator proposes to build a unified program representation that encapsulates high-level computational and data encodings, as well as low-level optimizations. An AI-driven ML compiler agent then interacts with structured tools to iteratively optimize the program. This project accelerates the deployment of advanced machine learning models across a broad spectrum of platforms, enabling rapid innovation and supporting the next wave of AI applications and toolchains. 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. NSF Award ID: 2543593 | Program: 01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Tianqi Chen | Institution: Carnegie Mellon University, PITTSBURGH, PA | Award Amount: $395,936 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2543593 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2543593.html