Development of Logistic Regression Algorithm for Graphics Processing Units

Artificial intelligence is getting more popular as it can solve a variety of problems much faster than humans, although training the model on a big dataset might take a lot of time. Sometimes utilizing Graphics Processing Units may be beneficial to accelerate an algorithm. However, for each algorithm that we want to use, we need to have an implementation that can effectively utilize specific GPU hardware. There are a few frameworks that provide implementations of the algorithms from classical machine learning that can run on NVidia-manufactured GPUs. Unfortunately, these implementations are vendor-specific and are not suitable for all types of devices. In this project, we introduce a new implementation of a logistic regression algorithm that is developed using SYCL standard, so it can utilize Intel-manufactured GPUs. We also provide a Python interface to make the use of this algorithm easier.