PhD Topics

Host: Eindhoven University of Technology; Academic partner: University College Cork; Industry partner: SANQUIN
Many real-world decision-making problems involve balancing multiple objectives. For example, Sanquin, the Dutch blood bank, aims to optimize its blood supply chain by reducing blood waste, lowering transportation costs, and ensuring fairness to donors. In this project, we will model such problems as multi-objective optimization problems and address them through a combination of deep learning, reinforcement learning, and algorithms. To this end, we will develop a general framework and computational tool that can be adapted to a variety of downstream optimization problems with well-balanced diverse objectives.

Duration: 4 years; Salary and benefits (such as a pension scheme, paid pregnancy and maternity leave, partially paid parental leave) in accordance with the Collective Labour Agreement for Dutch Universities, scale P (min. €3,059 max. €3,881). This salary is in accordance with the MSCA Call 2025 regulations for Doctoral Researchers and will be paid from the relevant monthly gross allowances (gross living allowance € 4.010 per month; gross mobility allowance € 710 per month; gross family allowance € 660 per month, only if applicable).

Host: University of Vienna; Academic partner: University of Bologna; Industry partner: Hapag-Lloyd
The container shipping industry faces large-scale optimization challenges that must be solved on a regular basis. One of these is empty container repositioning, in which empty containers are moved to places where they are needed. Current techniques for solving such problems lack robustness in two key ways. First, they often do not consider stochastic aspects of the repositioning problem, resulting in solutions that have higher than expected costs or incorrectly reposition material in the network. Second, when these flow problems are re-solved, previous solutions are ignored, leading to difficulty in understanding and analyzing the solution for decision makers. This PhD project will address both flaws using a combination of Operations Research and Machine Learning techniques based on real-world data and challenges from consortium partner Hapag-Lloyd.

Duration: 3 years; Salary scheme B 1

Host: Tilde; PhD program: University of Latvia; Academic partner: University of Bologna
Large language models (LLMs) have achieved remarkable fluency and adaptability across languages for various tasks including machine translation. However, their reliability and controllability remain uncertain, especially when translations must adhere to external constraints—such as terminology, style, or domain-specific requirements. This doctoral project aims to fine-tune open weight LLMs for machine translation use cases allowing for integration of soft constraints for machine translation. The candidate will (1) design novel approaches for incorporating soft constraints into fine-tuned translation models, allowing flexible yet guided adaptation to user or domain requirements; (2) develop benchmark datasets and evaluation methodologies to assess constraint robustness across multiple translation scenarios; and (3) empirically evaluate the resulting models’ performance and stability across diverse use cases, including multilingual, low-resource, and domain-adaptive settings.

Host: University of Bologna; Academic partner: National University Corporation Okayama; Industry partner: Electrolux Italia
In recent years, automated testing in real-world or virtual environments has attracted considerable industrial attention, as a promising technology to speed up processes, reducing failures, reducing costs, and improve safety when autonomous agents are involved. The field of appliance design also stands to benefit from this approach, provided that some significant scientific challenges can be addressed. This project will explore techniques to address a few key decision making problems arising in automated testing, with an emphasis on 1) combining data-driven, physics inspired, and expert-derived knowledge; and 2) enabling decision making in black-box or grey box dynamic environments, possibly involving autonomous mobile agents.

Host: University College Cork; Academic partner: IT University of Copenhagen; Industry partner: Allianz
We will develop an approach for using constraint programming models as a technique for verifying deep neural networks, develop specific methods for classification problems and we will demonstrate the approach in at least one high-risk use-case, as specified by the European Union’s AI Act. The project will push the state-of-the-art in deep neural network verification using advanced constraints technology.

Host: IT University of Copenhagen; Academic partner: University of Vienna; Industry partner: Elsevier
Some research fields, including machine learning, are experiencing explosive growth in submissions to large conferences. Quality of peer review depends on the development of scalable methods for optimal allocation of the available reviewers to the submitted manuscripts. This PhD project focuses on developing (a) interpretable deep-learning-based methods for matching experts to cases based on semantic analysis of their publication history, and (b) hybrid assignment methods that satisfy the target quality and fairness constraints. The project is co-led by experts in natural language processing and discrete optimization, and it includes an industry secondment at Elsevier, ensuring hands-on experience with various peer review practices and workflows.

Host: University of Vienna; Academic partner: Eindhoven University of Technology; Industry partner: OPTANO
Automatically learning heuristics for solving combinatorial optimization problems allows for data-specific heuristics to be generated directly from data with little to no code or input necessary from domain experts. Current approaches use deep reinforcement learning, but these techniques generate black-box models that are hard if not impossible to interpret. This leaves decision makers clueless as to why a particular decision is chosen by a model. This PhD project will investigate methods for automatically learning interpretable heuristics, allowing decision makers to understand why the model recommends particular decisions. The project will focus on decisions in routing and scheduling applications, harnessing real-world data from the industrial partner OPTANO.

Duration: 3 years; Salary scheme B 1

Host: Eindhoven University of Technology; Academic partner: University of Vienna; Industry partner: Hapag-Lloyd
As part of this doctoral project, we will develop novel knowledge representation techniques, algorithms for human-in-the-loop optimization, and interactive tools that combine graphical and natural language interfaces to gather and incorporate expert knowledge. The doctoral candidate will also take part in secondments at (1) Hapag-Lloyd AG, where they will test early prototypes in real-world logistics scenarios, and (2) University of Vienna, to align and exchange ideas with related projects on data-driven optimization. This is a unique opportunity to contribute to foundational research with strong real-world relevance in collaboration with academic and industrial partners.

Duration: 4 years; Salary and benefits (such as a pension scheme, paid pregnancy and maternity leave, partially paid parental leave) in accordance with the Collective Labour Agreement for Dutch Universities, scale P (min. €3,059 max. €3,881). This salary is in accordance with the MSCA Call 2025 regulations for Doctoral Researchers and will be paid from the relevant monthly gross allowances (gross living allowance € 4.010 per month; gross mobility allowance € 710 per month; gross family allowance € 660 per month, only if applicable).

Host: Vrije Universiteit Brussel; Academic partner: University of Vienna; Industry partner: OMP
How can reinforcement learning solve (scheduling) problems faster and better using with graph neural networks? How can we transfer information between problem instances, while dealing with potential concept drift? How can we better understand the generalization of reinforcement learning methods for solving optimization problems? If these questions excite you, we invite you to apply for this PhD topic focusing on explainable, across-instance learning.

You will be part of a multidisciplinary team and will have the opportunity to participate in planned secondments at (1) the University of Vienna to generalize the knowledge transfer methods across different domains and (2) OMP to apply and refine your research in an industry environment, contributing to developing a public benchmark using an optimization problem from OMP.

Duration: 4 years, with the first 3 years funded in accordance with the MSCA 2025 regulations: a tax-free PhD bursary €4010/month including social security contributions which will be deducted from this amount, a mobility allowance: €710/month, and a family allowance: €660/month, if applicable. The 4th year is funded by VUB under national regulations (FWO PhD scale). VUB offers a standard benefits package, including hospitalization insurance, access to sports and library facilities, and training opportunities.

Host: University of Lleida; Academic partner: University College Cork; Industry partner: OMP
The Satisfiability (SAT) problem, the first problem shown to be NP-complete, asks whether a Boolean formula in Conjunctive Normal Form has a satisfying assignment and is fundamental to computer science. SAT solvers have improved dramatically over recent decades due to techniques such as non-chronological backtracking and conflict-driven clause learning (CDCL), allowing modern solvers to handle industrial instances with hundreds of thousands of variables and millions of clauses.

More powerful SAT solvers can also improve Maximum Satisfiability (MaxSAT), the optimization version of SAT, which seeks to satisfy as many clauses as possible while distinguishing between hard and soft constraints. SAT-based MaxSAT algorithms repeatedly invoke a SAT solver to test bounds on the optimum. This thesis conjectures that significant progress in core-guided MaxSAT solvers (a family of MaxSAT solvers) requires selecting effective core sequences rather than merely optimizing search within a fixed sequence. Different core sequences can lead to vastly different performance, including exponentially harder instances, making their avoidance essential. In this thesis, we aim to design meta-algorithms that efficiently explore and select favorable core sequences.

Finally, gadgets allow us to transform a SAT instance into a Max2SAT instance for use with MaxSAT solvers. The combination of such gadgets with MaxSAT algorithms can yield proof systems stronger than Resolution. In this thesis, we aim to build on these efforts by offering new insights into generating SAT-to-Max2SAT gadgets that can be exploited more efficiently by MaxSAT solvers.

To demonstrate impact beyond benchmarks, the project will apply these techniques to challenging mathematical and industrial problems.

Host: Vrije Universiteit Brussel; Academic partner: University of Warwick; Industry partner: Cargoful
How can learning-based methods and constraint satisfaction approximate solutions for complex combinatorial allocations? How can we optimize the allocation of goods and services while considering social constraints? How can we incorporate sequential and uncertain decisions into non-stationary optimization? This doctoral project will focus on finding the answers to these questions.

Duration: 4 years, with the first 3 years funded in accordance with the MSCA 2025 regulations: a tax-free PhD bursary €4010/month including social security contributions which will be deducted from this amount, a mobility allowance: €710/month, and a family allowance: €660/month, if applicable. The 4th year is funded by VUB under national regulations (FWO PhD scale). VUB offers a standard benefits package, including hospitalization insurance, access to sports and library facilities, and training opportunities.

Host: University of Warwick; Academic partner: Vrije Universiteit Brussel; Industry partner: SANQUIN
Blood donation is a pillar of modern healthcare. But while people continuously provide for others, understanding how to make best use of the shared resources is also key. With uncertain and constantly changing demand from hospitals, we need AI tools to help us allocate the optimal amount and type of blood to different areas to maximise fairness. Supervised by machine learning and theory experts at Warwick and VUB Brussels, with an industry secondment at medical SANQUIN, the project wants to make breakthroughs in learning protocols for fair allocation of limited resources in highly uncertain and dynamic environments, such as blood donation markets.

Duration: 3 years; Monthly salary: £ 4,045.31 (increases to £ 4,464.84 if you opt out of the pension scheme). Additional family allowance if applicable.

Host: University of Warwick; Academic partner: Eindhoven University of Technology; Industry partner: Fundazione LINKS & Gruppo Torinese Trasporti
Whether it’s trains, ships or bits, our societies heavily rely on well-regulated transportation networks. But when channels have limited capacity, how do we decide who to give priority to? Delegating decision-support to AI means having rigorous models that are grounded on fairness. This PhD project is about laying the foundations of fair decisions and fairness-informed machine learning & optimization in the context of transportation networks. Supervised by game theory and machine learning experts at Warwick and Eindhoven, with a company secondment at the LINKS foundation and transportation company Gruppo Torinese Trasporti, the research will design tools for multiobjective optimisation in transportation networks to achieve optimal solutions in practice.

Duration: 3 years; Monthly salary: £ 4,045.31 (increases to £ 4,464.84 if you opt out of the pension scheme). Additional family allowance if applicable.

Host: University of Bologna; Academic partner: Eindhoven University of Technology; Industry partner: University Hospital of Bologna IRCCS, Sant’ Orsola Polyclinic

Queue management problems are widespread in the healthcare sector, arising from the management of constrained and critical resources such as workforce capacity, operating rooms, or intensive care units. One prominent example involves the management of organ transplantation programs, which involve making life-saving decisions and accounting for multiple challenges, including donor scarcity, tissue compatibility, and pervasive uncertainty, but also fairness considerations and regulation compliance. This PhD project will investigate the use of AI techniques, in particular Machine Learning and Constrained Optimization, for queue management problems in general, with an emphasis on those arising in organ transplantation problems. The research will be carried out in cooperation with the Sant’Orsola Research Hospital in Bologna.

Host: University College Cork; Academic partner: University of Bologna; Industry partner: University Hospital of Bologna IRCCS, Sant’ Orsola Polyclinic

We will develop an approach to building machine learning models that satisfy user-specified fairness objectives and we will extend the approach for scenarios where such models are used for decision support. Furthermore, we will demonstrate the approach in at least one high-risk use-case, as specified by the European Union’s AI Act. The project will create a suite of novel constraint programming and optimization techniques for building fairness-aware classification models and using them within decision support systems.