Open calls in doctoral studies, the fourth period

Supervisor(s): Sedat Akleylek

This doctoral research will investigate post-quantum cryptographic primitives and develop lightweight protocols for real-time applications, using power-system communication as the primary application domain. It will focus on key-encapsulation mechanisms and digital signatures as building blocks for secure key establishment, authentication and message integrity under strict computational, memory, bandwidth and timing constraints. The theoretical work will examine the security and efficiency of constructions combining post-quantum primitives, digital signatures, key agreement primitives. Particular attention will be given to signature generation and verification costs, signature and public-key sizes, and their effects on protocol performance. The research will develop methods for authenticated key establishment, key renewal and secure connection recovery, with explicit treatment of transcript binding, replay protection and downgrade resistance. Formal verification and, where feasible, computational security arguments will assess the proposed constructions under clearly defined adversarial assumptions. Analytical models will characterize security–performance trade-offs and guide protocol design. Selected implementations will be optimized for memory use, cryptographic processing and communication overhead while preserving standardized security parameters. Evaluation will combine embedded-platform benchmarks, network simulation and cyber-physical experiments aligned with SIPSIK’s power-system research. The intended innovation is the joint analysis and optimization of post-quantum primitives, digital-signature use, and protocol behavior for resource-constrained real-time environments. Expected outcomes include security-analyzed protocol constructions, efficient implementations, reproducible benchmarks and practical migration guidance. The research will establish when post-quantum authentication and secure communication can satisfy the operational requirements of power systems and related critical infrastructure.

Supervisor(s): Eduard Barbu

Large language models are increasingly used in systems that support human decision-making, yet it is often difficult to understand how a model arrives at a particular answer or decision. This doctoral thesis aims to investigate how such models can be made more transparent, interpretable, and controllable. The research will focus on open-weight language models, whose internal representations and computational processes can be directly analysed. The thesis will investigate which internal representations and signals are associated with particular model decisions and how interpretability methods, including feature attribution, representation analysis, counterfactual reasoning, and intervention methods, can be used to explain model behaviour. The research will also evaluate how faithful and stable these explanations are across different models, tasks, and application domains. An important part of the thesis will be the integration of technical interpretability methods with a conversational interface. The aim is to enable users to ask in natural language why a model made a particular decision, explore alternative scenarios, and receive explanations grounded in the analysis of the model's internal representations. The doctoral research will be carried out within the University of Tartu TEM-TA project “Interpretable AI with Open-Weight Language Models: A Conversational Framework for Transparent Decision Support.” The results are expected to contribute to the development of more transparent and trustworthy AI systems, particularly in settings where AI-supported decisions need to be understood, examined, and challenged.

Supervisor(s): Kallol Roy, Radwa Mohamed El Emam El Shawi
Muon tomography uses naturally occurring cosmic-ray muons to image dense or shielded objects without artificial radiation. Practical systems must work with a low muon flux, short exposure times, noisy measurements and a difficult inverse problem, and machine-learning models now used for object localisation and material recognition rarely show why they reach a decision. This doctoral project will develop interpretable machine-learning methods whose predictions of object location and material come with explanations that can be checked against individual muon tracks, scattering statistics and known physics. Building on the research group's dual-view analysis of straight and scattered muon rays, the project will study track-level attribution, explanations based on physical concepts and material prototypes, physically plausible counterfactuals and calibrated uncertainty. Explanations will be assessed quantitatively for faithfulness, stability, calibration and physical consistency, not only for visual plausibility. The methods will be tested first on controlled GEANT4 simulations and then on real detector data from the industrial partner GScan OÜ within the TEM-TA290 programme. The expected result is an auditable and robust machine-learning framework that explains why an object or material is detected, helps diagnose false, missed or merged objects, and supports trustworthy use of muon tomography in security inspection, non-destructive testing and infrastructure monitoring.

Supervisor(s): Prof. Marco Kirm, Assoc. Prof. Eduard Feldbach ja Assoc. Prof. Rainer Pärna

This doctoral thesis focuses on investigating the electronic, optical, and structural properties of nanostructured functional materials and on relating these properties to material functionality and application potential. The research objects include compounds with nanoscale structural features, including C12A7, apatites, and SiC-based materials. An important part of the work is the development of new synthesis methods aimed at producing materials with controlled composition, high purity, and desired phase composition in the form of powders, ceramics, and thin films.

The applied research direction of the doctoral thesis is related to the valorisation of silicon-rich waste from end-of-life crystalline silicon photovoltaic panels. The research will test the hypothesis that, through the targeted optimisation of process sequence, purification steps, and synthesis parameters, recovered silicon can be obtained at a quality suitable for use as a feedstock for the synthesis of higher-value-added SiC. This approach integrates circular economy principles, secondary raw material valorisation, and the development of advanced high-technology materials.

The scientific novelty of the work lies in the systematic investigation of the properties of nanostructured compounds, the development of new synthesis methods, and the integrated application of complementary characterisation techniques. XRD, XPS, SEM, micro-Raman, and luminescence spectroscopy, together with synchrotron-radiation-based analytical methods, will be used to determine the phase composition, microstructure, and electronic properties of the materials. The research is expected to generate new knowledge on structure–property relationships in functional materials and to support the development of higher-value-added materials, technological processes, and potentially patentable solutions.