Optimization research focuses on developing methods for solving difficult combinatorial and continuous optimization problems, with particular emphasis on the flexible job-shop scheduling problem, multi-objective optimization, and the analysis of fitness landscapes. A common thread is the use of machine learning—especially reinforcement learning and deep learning—within optimization frameworks.

Scheduling & Combinatorial Optimization

Flexible Job-Shop Scheduling with Deep Reinforcement Learning

An extensive research program on solving the flexible job-shop scheduling problem (FJSP) using deep reinforcement learning. Our work addresses the problem from multiple angles:

  • Graph representation: Novel graph representation enabling diverse policy generation via deep RL.
  • Real-time solving: Constraint programming integrated with deep learning for dynamic scheduling.
  • Offline RL: Learning scheduling policies from historical data without online environment interaction.
  • Self-evaluation: Methods for evaluating scheduling decisions without ground truth.
Behavioral Cloning for Scheduling
The Multi-Assignment Scheduler: a new behavioral cloning method for the job-shop scheduling problem. Learning scheduling heuristics from expert demonstrations using imitation learning.
Routing and Branching Strategies in 3D Manufacturing
Application of operations research and mathematical programming to routing problems in 3D design for manufacturing. Branching strategies for efficient exploration of the solution space.

Multi-Objective Optimization

Bi-Objective Combinatorial Optimization
Research on bi-objective combinatorial optimization problems with heterogeneous objectives (e.g., one continuous and one discrete objective). Development of algorithms for Pareto front approximation in such settings.
Benchmarking MOEAs for RL Problems
Systematic comparison of multi-objective evolutionary algorithms (MOEAs) for solving continuous multi-objective reinforcement learning problems. Analysis of NSGA-II, MOEA/D, and other state-of-the-art algorithms on RL benchmark problems.

Fitness Landscape Analysis

Analysis of Problem Structure
Research on characterizing the structure of optimization problems through fitness landscape analysis. Understanding why certain problems are hard for particular algorithms, and how landscape features relate to algorithm performance.
NK Landscapes & Problem Classes
Research on NK landscapes and related benchmark problem classes for understanding the difficulty of combinatorial optimization problems. Analysis of how problem structure affects the behavior of evolutionary algorithms.

Energy Systems Optimization

Geothermal Power Plant Optimization
Application of continuous estimation of distribution algorithms for parametric optimization of geothermal power plants. Optimizing operating parameters to maximize efficiency and reduce environmental impact.
Smart Grid Optimization
AI methods for smart electricity distribution grids: meter-transformer connectivity identification, voltage control, and anomaly detection in low-voltage grids.
Building Energy Management
Intelligent methodology for energy management in buildings and anomaly detection. Prediction of building energy consumption using evolved neural networks and other machine learning approaches.

Selected Publications