Medizinische Informatik, Statistik und Dokumentation

Research focus on Human-Computer-Interaction for Medicine and Healthcare

PI: Andreas Holzinger

Focus: Artificial intelligence has been remarkably successful due to advances in statistical machine learning, even surpassing human performance in certain tasks in medicine. However, the complexity of such approaches often makes it impossible to understand why an algorithm arrived at a particular result. The focus of our research is on comprehensibility and thus interpretability. This is where a human-in-the-loop can be helpful, because human experts can contribute experience, contextual understanding, implicit and conceptual knowledge.

Cooperation: The Research Team Holzinger cooperates intern with the Diagnostic and Research Institute of Forensic Medicine and international with the xAI-Lab of the Alberta Machine Intelligence Institute, Edmonton, the Life Sciences Discovery Center Toronto in Canada, and with the Human-Centered AI Lab at the University of Technology, Sydney, Australia.

Projects

AIDAVA - AI-powered Data Curation & Publishing Virtual Assistant

  • Work Package 5 of the AIDAVA project, particularly Task 5.3, focuses on developing a novel approach to explainability and human-AI interaction to maximize user acceptance. The aim is to develop tools to support annotators with human-in-the-loop (HITL) interfaces that combine human expertise with machine knowledge. The subtasks also include the identification of problems in explanation approaches, the definition of adaptive methods for different skill levels to maximize confidence, the definition and testing of HITL interface patterns, and the evaluation of the prototype.
  • Period: 2022-2026
  • Funded by: European Commission
  • Project partners: b!loba, KU Leuven, The European Institute for Innovation through Health Data, European Cancer Patient Coalition, European Heart Network AISBL, ONTO - Sirma AI EAD, NEMC - Sihtasutus Põhja-Eesti Regionaalhaigla, Averbis GmbH, European Research and Project Office GmbH, UM - Maastricht University, Egnosis by Gnome Design Srl, MIDATA Cooperative, Digi.me Ltda

Feature Cloud

  • As part of the EU-RIA project 826078 "Privacy preserving federated machine learning", where the goal is to exchange only learned representations (the feature parameters theta, hence the project name), the team works on distributed machine learning and contributes to explainability and interpretability of such approaches, in particular graph-based explainable AI and aspects of efficient human-AI interaction that support ethically responsible and legally defensible machine learning in medicine.
  • Duration: 2019-2024
  • Funded by: EU
  • Cooperation partners: TU Munic, Uni Hamburg, Uni Marburg, SBA-Research Vienna, University of South Denmark, Uni Maastricht, Research Institute Vienna, Gnome Design SRL

Principal Investigator

Andreas Holzinger  
T: +43 316 385 13883