Academic Awards 2025 booklet

81 The clinical standard to reduce dimensionality of PSG data yields a categorization of each 30-second multi-sensor window into one out of five discrete sleep stages. These sleep stages are distinguished according to a rulebased classification system – developed through expert consensus – that relies on visual inspection of characteristic signal features per sleep stage. Plotting the sleep stage annotations across the full night results in a visualization called a hypnogram. Due to the coarse categorization and the low temporal resolution of sleep staging, it creates an extremely compressed representation of PSG data and may, therefore, omit clinically-relevant information. Richer and less compressed representations might reveal new patterns in healthy sleep and improve our understanding, diagnosis and treatment of pathological sleep. The field of representation learning is concerned with finding a data-driven transformation that results in a – often low-dimensional – data representation that captures the independent factors of variation underlying a high-dimensional measurement. Asub-field focuses on discrete representations, where data are categorized in datadriven categories or clusters for the sake of additional dimensionality reduction, interpretability, or because the data are assumed to be inherently discrete. Modern discrete representation learning methods combine self-supervised feature learning with clustering approaches. The thesis of Iris focuses on the development of discrete representation learning methods for time series data, with the purpose of revealing sleep structure in PSG data. This work therewith paves the way to gaining a better understanding of both healthy sleep and disordered sleep which, in turn, can improve diagnosis and treatment of sleep disorders. This made Iris’ project very multidisciplinary and particularly challenging as it combined fundamental advances in machine learning with a clinical application area, requiring a profound knowledge and understanding of both fields. She published her work in an impressive number of journal and conference papers. Juliëtte van Haren Perinatal Life Support technology represents a future vision of life support technology that is at the cutting edge of the research being conducted in the field of extremely premature infant healthcare. This technology aims to extend organ maturation within an environment resembling the womb. After preterm birth, the infant would be kept in a fetal physiological state, with the lungs immersed in amniotic fluid, and a sustained fetal blood circuit through the use of an artificial placenta. The work of Juliëtte describes and investigates the technology from a perspective of Design, and not from medical ethics, clinical or engineering viewpoints. The work takes a patient-centered vision, to create solutions that not only meet medical and functional requirements but also address the needs and values of future patients, parents, relatives, clinicians, and other (in)direct users of the technology. The research is demonstrating that Design may be used to take on a patient-centered vision already in an early stage of research, and thereby contribute to safe and responsible development of high-risk technologies and enhance the overall quality of perinatal care. Juliëtte published her results in many journal papers.