Tensor Execution engine on GPU

My project proposal aims to optimize the FasTensor tensor computing library to work efficiently on GPUs to enable efficient tensor contraction while maintaining the structure-locality of tensor data. This project involves creating custom-defined computational operations on GPUs and is essential for many scientific applications, including advanced AI model training. My expected deliverables include a working implementation of FasTensor on GPUs, a report on the performance of the execution engine, and documentation of the execution mechanism. My implementation plan involves using a combination of programming languages, including C++ and CUDA, and managing memory and data movement efficiently. My project timeline includes research, planning, implementation, testing, documentation, and reporting. I have a relevant academic background and professional experience in software engineering and machine learning.