IREE Study on MLIR Based end to end compiler

Topics of this document : - Extension of an existing embedded ML runtime (IREE) to improve its timing predictibility, with a focus on memory allocation and scheduling. - Real-Time ML specification. Extend the internal formalism used by one ML framework (TensorFlow/MLIR) to allow the specification and manipulation of non-functional properties. - Real-time implementation of real-time reactive applications onto various execution platforms (GPUs, multi-cores ). - Use case implementation (focus on UC01 on Android-based and select SoCs) and integration of the tools within the Confidence Environment.