Machine Learning-Based Optimization Techniques for Modern Computing Applications

Abstract

The rapid proliferation of data-intensive computing applications has created an urgent demand for optimization techniques that can scale with modern system complexity. Machine learning (ML) has emerged as a transformative paradigm for solving optimization problems that are computationally intractable for classical algorithmic approaches. This paper presents a comprehensive survey of machine learning-based optimization techniques and their applications across cloud computing, network resource allocation, energy-efficient systems, and automated software engineering. We systematically categorize these techniques into gradient-based optimization, evolutionary and metaheuristic approaches, reinforcement learning-driven optimization, and hybrid frameworks that integrate ML with conventional optimization methods. For each category, we examine the underlying mathematical formulations, algorithmic architectures, and performance benchmarks reported in the recent literature. Furthermore, we analyze the practical deployment considerations including computational overhead, convergence guarantees, and generalizability across problem domains. Our analysis reveals that hybrid optimization strategies, which combine the exploration capabilities of metaheuristic algorithms with the learned representations of deep neural networks, consistently outperform standalone methods in multi-objective optimization scenarios. We also identify critical open challenges, including the need for interpretable optimization models, federated optimization for privacy-sensitive applications, and energy-aware optimization for edge computing environments. This survey provides researchers and practitioners with a structured reference for selecting and adapting ML-based optimization techniques to emerging computational problems.