OPTIMIZATION OF ARTIFICIAL INTELLIGENCE ALGORITHMS BASED ON MATHEMATICAL MODELS

Abstract

This article examines the optimization of artificial intelligence algorithms based on mathematical models. The relevance of the study is determined by the fact that the accuracy, generalization ability, computational cost, and stability of modern AI systems largely depend on the selected optimization methods. Based on the provided literature, the study applies theoretical analysis, comparative review, mathematical modeling, and conceptual synthesis. A bi-level model is proposed, combining empirical risk minimization, regularization, gradientbased parameter updates, and outer-loop hyperparameter optimization. As a result, analytical conclusions are drawn regarding the applicability of classical gradient methods, adaptive optimizers, and meta-heuristic approaches. The scientific novelty lies in the systematization of studies on AI optimization and in presenting them within a unified mathematical modeling framework.