Application of Machine Learning and Parallel Computing to Search for Hypersurfaces Containing Data in Non-Linear Spaces

To apply machine learning in practice, you often need to process a large amount of data and this takes a lot of time. The use of parallel computing and other types of parallelism can significantly speed up calculations, sometimes even without changing the algorithm itself. This paper explores the use of machine learning and parallel computing to search for hypersurfaces containing data in non-linear spaces. Since the transfer of data between the CPU and GPU is very speed- limited, it is important to be able to generate geometry directly on the GPU. Using various vertex arrays makes it easy to transfer a large amount of various vertex attributes (such as coordinates, normal/binormal, color, texture coordinates, etc.) directly to the GPU and quickly render data. The main advantages of using CUDA for parallel computing are its simplicity and flexibility.