This National Science Foundation (NSF) Project Grant award under the Computer and Information Science and Engineering (CISE) program, titled "COLLABORATIVE RESEARCH: CNS CORE: SMALL: A COMPILATION SYSTEM FOR MAPPING DEEP LEARNING MODELS TO TENSORIZED INSTRUCTIONS (DELITE)", provides $299,999 in funding to the University Of Georgia Research Foundation, Inc. over a 3-year period from October 1, 2023 through September 30, 2026.
The grant supports the development of a compilation system that can efficiently map modern deep learning models to emerging tensorized instruction sets across different hardware architectures. Key focus areas include local operator-level optimization, global model-level optimization, handling dynamic neural network models, and improving efficiency for sparse models. The system will take a computational graph representation as input and generate optimized tensor and LLVM IR outputs for deployment on a variety of hardware backends. This work aims to provide significant technical, economic, and societal impacts by improving the efficiency of AI/ML workloads on modern computing platforms.