Acquiring and preprocessing annotated PSG datasets from various sources using the open EDF standard, assessing inter-database differences, and establishing a baseline for human agreement.
Developing deep learning algorithms based on attention mechanisms to enable robust and efficient identification of multiple event types within PSG data.
Human-in-the-loop strategies, including Curriculum Learning, Active Learning, and Interactive Machine Learning, to enhance model training through expert feedback and dynamic data selection based on case difficulty.
Quantum machine learning for biomedical signals, focusing on data encoding, quantum model design, and integrating attention mechanisms to enable transformer-like architectures.
Integrate explainability into ML models using self-attention mechanisms, visualizing attention over PSG signals, and incorporating model-agnostic methods.
Integrate developed solutions into a scalable, generalizable model using heterogeneous datasets, testing ensemble strategies, and evaluating performance in clinical settings.
Technical reports and scientific papers.
Project reports and data management plans.
