Loading...
Machine Learning for Combinatorial Optimization
Iklassov, Zangir
Iklassov, Zangir
Files
Author
Supervisor
Department
Machine Learning
Embargo End Date
Type
Dissertation
Date
2024
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Combinatorial optimization problems (CPs) are crucial in logistics, resource management, and scheduling. Traditional methods such as exact algorithms, heuristics, and metaheuristics often struggle in large or dynamic environments. This thesis explores machine learning approaches, focusing on neural combinatorial optimization and reinforcement learning (RL), to address these limitations. Key contributions include a curriculum learning strategy for job shop scheduling, which improves model performance by dynamically adjusting task difficulty, and an RL framework for stochastic vehicle routing, which effectively handles uncertain demands. Additionally, we introduce an RL framework for time-constrained vehicle routing and a self-guiding exploration strategy for large language models (LLMs) to enhance combinatorial problem-solving. These innovations demonstrate the potential of machine learning to significantly improve the efficiency and adaptability of CP solutions.
Citation
Z. Iklassov, "Machine Learning for Combinatorial Optimization", PhD Dissertation, Machine Learning, MBZUAI, Abu Dhabi, UAE, 2024