Shayan Dadman disputerer for ph.d.-graden i ingeniørvitenskap / Shayan Dadman will defend his thesis for the PhD degree in Engineering Science
Shayan Dadman disputerer for ph.d.-graden i ingeniørvitenskap og vil offentlig forsvare avhandlingen / Shayan Dadman will defend his thesis for the PhD degree in Engineering Science:
“Beyond Autonomous Creation in AI Music Generation: Multi-Agent Perception-Policy Separation with Process-Oriented Evaluation for User-Centric, Adaptive Systems”.
Avhandlingen er tilgjengelig her (lenke kommer) / The doctoral thesis is available here (the link will be available soon).
Auditoriet er åpent for publikum. Disputasen vil også bli strømmet. Opptak av disputasen vil være tilgjengelig i en måned. / The auditorium is open to the public. The defense will be streamed. A recording of the defense will be available for one month.
Prøveforelesningen starter kl. 10:15 / The trial lecture starts at 10:15. Tittel:
“***”.
Disputasen starter kl. 12:15 / The defense starts at 12:15.
Prøveforelesning strømmes her, disputas strømmes her / The trial lecture will be streamed here, and defense will be streamed here.
Sammendrag av avhandlingen / Summary of the thesis:
Contemporary AI music generation systems demonstrate remarkable autonomous capabilities, yet remain limited in supporting interactive human-AI collaboration. This thesis investigates whether such limitations represent engineering deficits or architectural constraints requiring structural intervention. Systematic analysis of 73 deep learning systems reveals that approximately 85% offer only pre-generation control, while fewer than 15% support during-generation interactivity—indicating architectural rather than resource-based limitations. In response, this research advances Perception-Policy Separation as a foundational architectural intervention. By decoupling perceptual organisation from generative decision-making, explicit pathways emerge for real-time human guidance without requiring model retraining. The thesis also contributes SCOPE, a process-oriented evaluation framework assessing workflow integration across goal formation, exploration, and refinement phases—addressing gaps in existing evaluation methods. Together, these contributions establish both design criteria and evaluation methodology for adaptive music generation, providing foundations for extending interactive AI capabilities across broader creative domains.
Hovedveileder / Main supervisor:
Prof. Bernt Arild Bremdal, Department of Computer Science and Computational Engineering, Faculty of Engineering Science and Technology, UiT The Arctic University of Norway.
Biveileder / Co-supervisor:
Prof. Rune Dalmo, Department of Computer Science and Computational Engineering, Faculty of Engineering Science and Technology, UiT The Arctic University of Norway,
Prof. Børre Bang, Department of Computer Science and Computational Engineering, Faculty of Engineering Science and Technology, UiT The Arctic University of Norway.
Bedømmelseskomité / Evaluation committee:
Prøveforelesning og disputas ledes av prodekan for forskning, Prof. Bjarte Hoff / The trial lecture and defense are led by Vice Dean of research, Prof. Bjarte Hoff.
De som ønsker å opponere ex auditorio kan sende e-post til Bjarte Hoff / Opponents ex auditorio should contact Bjarte Hoff.