Using OpenMP to Optimize Model Training Process in Machine Learning Algorithms

It is obvious the digital world is continuing to expand while the completion of Big Data solutions frequently relies on the timely extraction of valuable insights from data. This will continue to become more challenging due to the rise in data volume without a corresponding increase in velocity. Therefore, it has the potential to optimize storage space, decrease the amount of processing required for further information extraction, and save I/O and network communications. In Artificial intelligence specifically, implementing machine learning algorithms on big data, such as social networks and web graph, is challenging. Even though many types of researches have focused on making sequential algorithms more scalable, however, their running times remain to be prohibitively long. It becomes necessary to divide the work among multiple threads. In this paper, we have used the power of parallel computing particularly OpenMP implementation in supervised machine learning processes to reduce computation time. An experiment was done on a binary classification problem, that was trained on a set of realistic data and it proved a significant reduction in calculation time compared to classical algorithms. In the end, we have described possible directions for further researches concerning parallel optimization of time calculation in the supervised perceptron learning processes.