Reading time ~3 minutes
Optimization is a cornerstone of problem-solving in numerous scientific, engineering, and industrial fields. Many real-world challenges involve identifying the most efficient solution within complex constraints, where traditional methods often fall short. Our research focuses on developing and applying hybrid optimization techniques, particularly those integrating genetic algorithms (GAs) and biased random-key genetic algorithms (BRKGA) with reinforcement learning, to solve practical, real-world problems across various domains.
The papers below are a selection; the full list is on my Publications page.
Hybrid Methods: Combining Strengths for Superior Solutions
Genetic algorithms are widely recognized for their ability to explore vast search spaces and identify near-optimal solutions through evolutionary principles. However, when applied in isolation, they can face challenges such as slow convergence, premature stagnation, and difficulty refining local optima. To overcome these limitations, we focus on hybrid approaches that combine GAs with complementary optimization techniques.
Our hybrid methods integrate GAs with:
- Local search algorithms – for fine-tuning solutions and accelerating convergence.
- Machine learning models – to guide search processes and adapt optimization strategies dynamically.
- Heuristic and metaheuristic methods – to balance exploration and exploitation more effectively.
- Problem-specific refinements – incorporating domain knowledge to improve optimization efficiency.
By leveraging the strengths of multiple optimization paradigms, our hybrid methods provide robust, scalable, and adaptive solutions that outperform conventional approaches in both speed and accuracy.
Application Areas: Solving Real-World Challenges
Our research is driven by the need to apply advanced optimization techniques to real-world, high-impact problems. Below are the key areas where our hybrid optimization approaches have been successfully implemented, each backed by concrete studies.
Scheduling and Resource Allocation
Optimizing limited resources for maximum efficiency in logistics, workforce planning, supply chain management, and combinatorial problems where good schedules or allocations are hard to find.
Selected publications
- “A Hybrid Approach with BRKGA and Data Mining for the Early/Tardy Scheduling Problem” — CEC 2024
- “A Q-Learning Hybrid BRKGA Applied to the Knapsack Problem with Forfeits” — CEC 2024
- “A Hybrid BRKGA Approach for the Multiproduct Two Stage Capacitated Facility Location Problem” — CEC 2022
Hyperparameter and AI Model Optimization
Fine-tuning machine learning hyperparameters and architectures—one of the most expensive steps in modern ML—by guiding the search with evolutionary and reinforcement-learning-based strategies.
Selected publications
- “A Population-based Hybrid Approach for Hyperparameter Optimization of Neural Networks” — journal
- “Efficient Hyperparameter Optimization Using Deep Q-Network and BRKGA” — CEC 2024
- “A conjugated evolutionary algorithm for hyperparameter optimization” — WCCI/CEC 2022
Network and Graph Optimization
Finding optimal configurations for communication and sensor networks, and reducing the size of large graphs so that downstream models run faster without losing accuracy.
Selected publications
- “Optimizing Wireless Sensor Network Topology With Deep Reinforcement Learning for Multi-Source/Destination Scenarios” — BDCAT 2024 (Best Paper Award)
- “Efficient Node Reduction Heuristic for GNN-Based Traffic Speed Forecasting” — BDCAT 2024
- “A Multi-centrality Heuristic for the Bandwidth Reduction Problem” — ICCSA 2024
Optimization for Real-World Operations
Bringing metaheuristics to bear on operational problems with direct social and environmental impact, such as emergency and rescue logistics.
Selected publications
- “ACO With Reinforcement Learning Applied to Rescue Operations on Urban Forests” — CEC 2024
Our practical, problem-driven approach ensures that the solutions we develop are not only theoretically sound but also directly applicable to real-world challenges.
Future Directions and Collaboration Opportunities
Optimization is a continually evolving field, and we are always exploring new ways to enhance hybrid optimization techniques. Our ongoing research aims to:
Improve the efficiency and scalability of hybrid methods for complex problems. Explore adaptive and self-learning optimization frameworks. Investigate new applications in emerging fields such as AI-driven automation, smart systems, and sustainable resource management. We welcome collaborations with researchers, industry professionals, and students interested in advancing hybrid optimization techniques. If you are working on a complex optimization problem and are looking for innovative, data-driven solutions, we are eager to explore potential research partnerships.
Questions?
Have some question about it? You can send me an email