Stochastic Relational AI in Healthcare

Intelligent systems that build models by observing their environment and evaluating data to calculate actions optimally must systematically deal with uncertainties. Especially in healthcare, diverse and to a large extent relational data of patients, their medical history, and their examinations with measurement series and diagnoses are encountered. Associated text and image data are often to be interpreted on a semantically higher level and made usable for novel applications.

Stochastic relational models represent a crucial tool here. They are also of great importance in other areas such as robotics or the Semantic Web, since extensive data sets from web mining activities (e.g., so-called knowledge graphs) are increasingly being included in the action planning of intelligent systems. Through Stochastic Relational AI in Healthcare (StarAI), stochastics/statistics, logic, and network models are systematically combined for use in intelligent systems. For example, important problems can be solved in health care system applications. Therefore, the scalability of inference and learning algorithms for StarAI models forms an important work topic in the research department.

The human-friendly design of the interaction of intelligent systems with human actors receives special attention. An essential prerequisite here is a very high quality of system outputs or a very high appropriateness of computed actions or recommended actions of intelligent support systems based on possibly learned models. Equally important are the possibilities to explain conclusions to the user with reference to the modeling with causal reference, but also to anticipate human information needs as well as human information processing in the interaction. Both aspects are investigated in the research area (Human-aware StarAI).

StarAI uses the basic research of the Institute for Information Systems (IFIS) at the University of Lübeck, led by Prof. Dr. Ralf Möller.

IFIS projects with StarAI reference

• Prof. Möller is the speaker of the research unit Data Linking in the DFG Cluster of Excellence Understanding Written Artefacts. In this context, there are several subprojects.
• The project KI-Lab (BMBF) is building an AI research infrastructure at the University of Lübeck.
• The Mittelstand 4.0 Competence Center Kiel (BMWi) supports SMEs in mastering digital transfomation.
• The Optique project (EU) investigated ontology-based access to stream-based data along with very large relational datasets. Optique has contributed to the semantics of query languages and the efficiency of algorithms to execute queries for continuous prediction of events in high-speed data streams and historical data.
• The PANOPTESEC project (EU) investigated monitoring techniques for cybersecurity in critical infrastructures.
• The Stochastic Relational in AI and Healthcare Big Data (CISCO) project has laid the groundwork for a remote monitoring system of patients without adequate medical care.