The list of topical topics of thesis topics is Master's degree. Topics are grouped by supervisors. The student can choose the proposed topic or agree his own with the supervisor.
Bychkov Oleksii Serhiiovych
Implementation of the scheduler based on the Pyhop/SHOP2 library with support for (PN)prerequisites. Integration with swarm simulator. Experiments in scenarios with conflicting information from different agents.
Finalization of the Question Engine module of the Dr.Case system (v2.0). Replacement of empirical bonuses ×10/×5/×2 with calibrated ones from the theory of possibilities. Experimental evaluation of improvement of diagnostic metrics.
Extension of the current Dr.Case NLP module (≥100 synonyms) to a full-fledged NER for medical terms in Ukrainian. Fine-tuning XLM-RoBERTa or UkrLaBSE. Integration with symptom_extractor.
Frontend on React/Next.js to conduct a consultation: the doctor introduces the case → specialist agents issue a hypothesis → visualization agreement → final conclusion. WebSocket for real-time updates.
Development of iOS/Android application with built-in Dr.Case model (model < 50 MB via quantization). Synchronization with the server when reconnecting. Ukrainian localization.
Development of a prototype of a multi-agent version of Dr.Case: specialist agents (therapist, cardiologist, neurologist) exchange hypotheses and agree on the diagnosis. Implementation of a negotiation protocol based on (PN)models.
A prototype implementation of multi-agent RL for distributed diagnostics, where agents cooperatively decide to whom to ask a clarifying question. Experiments on a synthetic case.
Implementation of CNN based on transfer learning (ResNet/DenseNet) for the detection of pneumonia, tuberculosis and neoplasms on radiographs. Probability calibration. rest integration with Dr.Case as the 19th module. Experimental evaluation on public datasets (CheXpert, NIH ChestX-ray14).
Implementation of a PBFT/Raft variant adapted to high mobility and partial channel losses. Experimental study in NS-3 or ROS 2 simulator.
Development of a system for simultaneous localization and mapping for a group of UAVs with different sensors. Experimental assessment of accuracy in environments without satellite navigation.
Finalization of the v2.0 module of the Dr.Case system with the replacement of heuristic likelihoods (0.90/0.10/0 .80/0.20, etc.) with calibrated ones according to the data. Experiments with ELC (Expected Likelihood Calibration).
Development of an ML model for the interpretation of CBC (leukocytes, erythrocytes, platelets, hemoglobin) with the detection of pathognomonic combinations. Integration as a Dr.Case module. Bayesian neural networks to assess uncertainty.