Parallel computing on GPU devices, alongside Machine Learning, has emerged as a key area of focus due to the substantial performance gains it offers. This integration of machine learning with GPUs presents numerous advantages, including accelerated computation speeds, enhanced parallel processing capabilities, and improved efficiency in handling large datasets. This synergy facilitates more sophisticated algorithms, enabling real-time data analysis and model training, thereby driving advancements in various applications ranging from deep learning to complex simulation tasks. In order to effectively understand the scope of this field, we have conducted a visualized bibliometric analysis utilizing VOSviewer and Biblioshiny. This analysis aims to elucidate the research trends in GPU-accelerated machine learning algorithms, based on articles indexed in Scopus, published between 1999 and April, 2026. The study indicated that a total of 1317 documents were published in 772 journals. The annual growth rate of the literature has been identified as 12.49% over the past two decades. Two of the highest collaborating countries are China and the USA, with China contributing almost 27% of the total publications. NVIDIA has contributed the most number of documents in this area of study. Furthermore, Deep learning algorithms significantly leverage the computational capabilities of Graphics Processing Units (GPUs) through advanced parallelization techniques. Moreover, image processing algorithms are among the most extensively parallelized workloads executed on GPUs, capitalizing on their architecture to enhance performance and efficiency. Although initial advancements in the literature were gradual, the advent of GPUs for general-purpose computing, alongside programming models like CUDA and frameworks such as TensorFlow and PyTorch, has markedly broadened the scope for research in the field.
Bidye et al. (Thu,) studied this question.