Explore the programs and courses offered by PhD in Computer Science specializing in AI and Operational Research
Browse Programs Admission InformationThe doctoral training program in computer science, offered by the Faculty of New Information and Communication Technologies at the University of Constantine 2 Abdelhamid Mehri, aims to train a new generation of highly qualified researchers capable of innovating and addressing national and global challenges. Aligned with national strategic priorities—such as citizen health, energy security, and food security—the program adopts a multidisciplinary approach focusing on artificial intelligence, natural language processing (NLP), machine learning, deep learning, cybersecurity, ambient intelligence, and nanotechnologies.
Supported by renowned research laboratories like TAMAYOUZ MISC, LIRE, and LISIA, and in partnership with national socio-economic stakeholders, the program aims to train 20 doctoral candidates capable of conducting high-level applied and theoretical research, while promoting innovation, interdisciplinary collaboration, and societal impact.
The first year of the doctoral program includes a set of fundamental courses spread over two semesters, aimed at strengthening the scientific, technical, and methodological skills of doctoral candidates. These courses cover artificial intelligence, advanced computer systems, cybersecurity, big data processing, and the foundations of scientific research.
Doctoral candidates also participate in mandatory seminars on research methodology, innovative project management, scientific communication, and ethical issues related to technology. Specialized workshops and optional seminars are offered to allow students to tailor their training path according to their research areas.
Proposal Title:
Hybridizing Reinforcement Learning and Traditional Optimization for Combinatorial Problems
Description: Combinatorial optimization problems, such as the traveling salesman problem, vehicle routing, and job scheduling, are central to operations research and computer science. These problems are notoriously difficult due to their discrete, high-dimensional solution spaces and NP-hard complexity. While traditional optimization techniques—such as branch-and-bound, dynamic programming, and heuristic/metaheuristic algorithms—have been effective in many domains, they often struggle with scalability or require significant domain-specific tuning.
This PhD research aims to explore a hybrid framework that integrates Reinforcement Learning (RL) with traditional optimization techniques to enhance solution quality, adaptability, and computational efficiency for combinatorial problems. The central hypothesis is that RL, with its ability to learn effective decision-making policies through interaction with the environment, can complement classical methods by guiding search heuristics, learning instance-specific strategies, or dynamically adjusting algorithmic parameters during optimization.
The project will focus on:
Designing architectures where RL agents interact with and enhance classical optimization procedures.
Investigating different hybridization strategies, such as RL-guided construction heuristics, local search improvements, or controller policies for metaheuristics.
Developing benchmark datasets and conducting extensive empirical evaluations on canonical combinatorial problems.
Analyzing theoretical properties of the hybrid approaches, including convergence and generalization.
The expected contributions include novel hybrid algorithms, theoretical insights into the synergy between learning-based and rule-based methods, and practical tools for solving large-scale, real-world combinatorial problems more effectively.