Mathematical modeling of neural network compression techniques for mobile platforms
Ibragim Mamadaev, Alina Mazhitovna Minitaeva · Journal of Physics Conference Series · 2025
Abstract Mathematical modeling plays a critical role in the optimization of neural networks, particularly when adapting machine learning algorithms for resource-constrained mobile platforms. This paper explores neural network compression techniques, including model quantization, pruning, and computational graph optimization, through the lens of mathematical modeling. By developing formal representations of these methods, we analyze their impact on performance metrics such as execution speed, memory usage, and accuracy on mobile devices. Additionally, we examine how mathematical frameworks can guide the integration of hardware-specific optimizations, such as leveraging Apple’s Neural Engine and Metal Performance Shaders. This study demonstrates that combining mathematical modeling with empirical evaluation leads to effective strategies for improving the efficiency of machine learning classification models on mobile platforms.