Research Area C | Domain-specific Machine Learning
C1 | Medicine
As part of the scientific goals, the research in MCML in Medicine and Healthcare will focus on research objectives that are necessary to overcome the hurdles for the deployment of ML approaches in clinical environments. In particular, advances are required in interpretable and explainable deep learning, robust and data efficient learning, privacy preserving learning, as well as in trust and safety of autonomous AI and ML systems.
C2 | Biology
MCML strategic aims in Biology and Biomedicine are to remedy current deficits of ML/AI research of particular importance in biomedicine: AI liability, blackbox behavior and explainability, privacy and security, and robustness; to enable and support revolutionary developments such as personalized drug development, smart nutrition, personalized medication, better diagnosis, personalized disease prevention, realtime health analytics, and personalized therapy; to combine the power of algorithms and human experts –scientists and medical doctors– to define new modes of biology/biomedical research as well as clinical patient care. MCML aims at being the interface for medical, biology, systems biology, and bioinformatics curricula and PhD programs to train the next generation of AI empowered scientists, clinical researchers, and medical doctors.
C3 | Physics and Geo Sciences
Geoinformation derived from Earth Observation satellite data is indispensable for tackling grand societal challenges. As part of the scientific goals, the research in MCML in Earth Observation will focus on developing and tailoring data science and ML concepts for georelevant application, taking the special characteristic into account. Specific research directions are physics-aware ML that generates models learned with deep neural networks, quantifications of uncertainty, explainable geoinformation retrieval, Quantum ML as a key future technology for a digital twin of the Earth, and ethical accompanying ML for Earth Observation.
C4 | Computational Social Sciences
The world has changed for empirical social scientists. New types of data from digital traces, new forms of access to traditional data, broad coverage of entire social networks, and computational advances to simulate social processes have shaped what is often referred to as computational social science. Our researchers are addressing the problem that in order to successfully use digital trace data, the research goals must be aligned with the available data and measurements. Data quality must be evaluated with the research goal in mind. To ensure reproducibility and replicability, documentation of digital trace data collection and processing is necessary. AI systems – even those designed with the best intentions – can be biased, operate unfairly, and increase social inequality if data feeding into the systems are not carefully assessed and evaluated.
LMU München
Statistics and Data Science in Social Sciences and the Humanities
C5 | Humane AI
MCML assumes that research in ML and data science aims at providing value for humankind. Innovations in ML must provide a benefit to humans, either for the individual or for society. Goals include making human actions more efficient, simplifying tasks or automating them completely, and uncovering insights and knowledge. Basic research in ML, such as optimizing algorithms or the invention of new methods, is typically generic and not linked to a specific application. However, this does not limit their value; it is rather the opposite, as basic research can provide benefits across application domains and for different purposes. Our basic research in human-centered ML will bridge this gap by eliciting requirements through human-centered methods and by envisioning, creating, and evaluating methods and technologies for efficient human-algorithm-data interaction. This is based on human-computer interaction, but moves beyond the computer as an artifact and towards interaction with data and intelligent algorithms and systems.
