Benchmarking ML/DL libraries

Abstract

The exponential growth of machine and deep learning applications across diverse domains has led to the proliferation of machine and deep learning libraries. As these libraries offer various algorithms, tools, and functionalities, choosing the right one for a particular task becomes pivotal. This project presents a comprehensive benchmarking study that systematically evaluates and compares the performance and usability of prominent machine learning libraries. The research methodology involves designing standardized tests covering a spectrum of tasks, including image classification, sentiment analysis, and regression. The study investigates libraries such as TensorFlow, PyTorch, Scikit-learn, and CuML, analyzing their efficiency, computational requirements, ease of use, and flexibility. The findings aim to assist practitioners, researchers, and developers make informed decisions when selecting the most suitable machine-learning and deep-learning library for their specific requirements.