PhD Defence: Investigating Implementation Issues of Quantum Machine Learning
On Friday, 19 December 2025, M.Sc. Ilmo Salmenperä defends his PhD thesis, "Investigating Implementation Issues of Quantum Machine Learning". The defence starts at 13:00 in the University of Helsinki Main Building, Auditorium Karolina Eskelin (U3032, Unioninkatu 34, 3rd floor). The defence will be held in English.
The thesis is part of ongoing research in the Department of Computer Science and the Empirical Software Engineering group at the University of Helsinki.
Event details
- Time: Friday, 19 December 2025, 13:00
- Venue: Auditorium Karolina Eskelin (U3032), University of Helsinki Main Building, Unioninkatu 34, 3rd floor
- Opponent: Professor Ilkka Tittonen (Aalto University)
- Custos: Professor Jukka K. Nurminen (University of Helsinki)
- Supervisor: Professor Jukka K. Nurminen (University of Helsinki)
- Language: English
What the thesis studies
Quantum Machine Learning (QML) promises practical gains by combining quantum computing with established machine learning methods. Turning that promise into working systems exposes two classes of issues: constraints from current hardware and challenges that arise from how models are formalized.
This thesis analyzes both, proposes solutions in concrete settings, and distills patterns researchers and engineers can apply across QML work.
Hardware-focused findings
Physical limits-qubit counts, gate topology, and error rates-bound the size and shape of problems we can run. Results show that careful restructuring of models often makes them viable on today's machines.
- Gate-based approaches: Quantum compilation routines were used to optimize an existing quantum linear regression method so it runs on current hardware.
- Quantum annealing: Classical software techniques improved the efficiency of embedding a Restricted Boltzmann Machine onto an annealing device.
Model-construction findings
Some issues originate in the model itself. Choices in formalization can trigger unintended effects, which can be addressed either by adapting usage or refining the model.
- Self-erasing gates: Identified in common constructions of quantum embedding kernels and mitigated with adjusted designs and usage guidelines.
- Feature permutation sensitivity: The order of features can significantly affect the performance of variational QML models; the thesis proposes practices to reduce that sensitivity.
Why it matters
The work spans multiple QML models and hardware families. It provides practical guidance for building algorithms that actually run, bridging the gap between theory and current devices.
For readers new to the area, see concise overviews of Quantum Machine Learning and Quantum Annealing.
Access to the dissertation
An electronic version will be made available in the University of Helsinki's open repository, Helda. Printed copies will be available on request from Ilmo Salmenperä.
About the research environment
The thesis is conducted within the Department of Computer Science and the Empirical Software Engineering group at the University of Helsinki, with supervision by Professor Jukka K. Nurminen.
For applied AI upskilling
If you are mapping skills for adjacent AI workstreams, explore current programs here: Latest AI courses.
Your membership also unlocks:
