Research on High-Dimensional Matrix Multiplication in Machine Learning for MEC

Machine learning (ML) algorithms have been widely deployed in mobile edge computing (MEC). Howerver, algorithms in MEC always demands high computational power to process complex models and large datasets in real-time. A critical component of this computational workload is matrix multiplication, a fundamental operation in many ML algorithms, particularly in deep learning. To meet the stringent latency and performance requirements of MEC applications, leveraging Graphics Processing Unit (GPU) acceleration for matrix multiplication is essential. GPUs, with their massive parallel processing capabilities, can significantly enhance the efficiency and speed of matrix computations, enabling rapid inference and training at the edge. This acceleration not only reduces the latency but also offloads the central cloud, providing a more distributed and scalable solution for real-time ML applications at the edge. To this end, this paper first introduces the background of using matrix multiplication to accelerate model training speed in ML for MEC, clarifies the significance of accelerating matrix multiplication to ML and MEC, then reviews and experiments on existing schemes for accelerating matrix multiplication and compares and summarizes various schemes from the perspective of different sparsity of matrices on the GPU. Finally, we points out the possible future research directions and technical challenges, and provides the certain reference value for the research in related fields.