This project focuses on advancing differentiable programming (DP) for efficient mapping of image processing and computer vision algorithms onto heterogeneous hardware (consisting of CPU, GPU en FPGA) and edge devices. The research aims to extend DP with techniques like expectation propagation and variational approximations, enabling distributed computing and gradient computations across cloud and edge platforms. The potential of FPGA-GPU hybrid systems for real-time video processing and the application of approximative computing techniques on edge devices is also explored, with an emphasis on reducing power consumption and improving latency (reduction and stabilization). The main goal is to support distributed inference and training for both neural networks and hybrid AI-image processing algorithms on diverse hardware architectures.