Der Abschlussbericht beschreibt Aufgaben und Ergebnisse des KI-Labors "ITSE". Gegenstand des KI-Labors bildeten Methodik, Technik und Ausbildung in der IT-Systemtechnik zur Analyse, Planung und Konstruktion KI-basierter, komplexer IT-Systeme.

While patients are known to respond differently to drug therapies, current clinical practice often still follows a standardized dosage regimen for all patients. For drugs with a narrow range of both effective and safe concentrations, this approach may lead to a high incidence of adverse events or subtherapeutic dosing in the presence of high patient variability. Model-informedprecision dosing (MIPD) is a quantitative approach towards dose individualization based on mathematical modeling of dose-response relationships integrating therapeutic drug/biomarker monitoring (TDM) data. MIPD may considerably improve the efficacy and safety of many drug therapies. Current MIPD approaches, however, rely either on pre-calculated dosing tables or on simple point predictions of the therapy outcome. These approaches lack a quantification of uncertainties and the ability to account for effects that are delayed. In addition, the underlying models are not improved while applied to patient data. Therefore, current approaches are not well suited for informed clinical decision-making based on a differentiated understanding of the individually predicted therapy outcome. The objective of this thesis is to develop mathematical approaches for MIPD, which (i) provide efficient fully Bayesian forecasting of the individual therapy outcome including associated uncertainties, (ii) integrate Markov decision processes via reinforcement learning (RL) for a comprehensive decision framework for dose individualization, (iii) allow for continuous learning across patients and hospitals. Cytotoxic anticancer chemotherapy with its major dose-limiting toxicity, neutropenia, serves as a therapeutically relevant application example. For more comprehensive therapy forecasting, we apply Bayesian data assimilation (DA) approaches, integrating patient-specific TDM data into mathematical models of chemotherapy-induced neutropenia that build on prior population analyses. The value of uncertainty quantification is demonstrated as it allows reliable computation of the patient-specific probabilities of relevant clinical quantities, e.g., the neutropenia grade. In view of novel home monitoring devices that increase the amount of TDM data available, the data processing of sequential DA methods proves to be more efficient and facilitates handling of the variability between dosing events. By transferring concepts from DA and RL we develop novel approaches for MIPD. While DA-guided dosing integrates individualized uncertainties into dose selection, RL-guided dosing provides a framework to consider delayed effects of dose selections. The combined DA-RL approach takes into account both aspects simultaneously and thus represents a holistic approach towards MIPD. Additionally, we show that RL can be used to gain insights into important patient characteristics for dose selection. The novel dosing strategies substantially reduce the occurrence of both subtherapeutic and life-threatening neutropenia grades in a simulation study based on a recent clinical study (CEPAC-TDM trial) compared to currently used MIPD approaches. If MIPD is to be implemented in routine clinical practice, a certain model bias with respect to the underlying model is inevitable, as the models are typically based on data from comparably small clinical trials that reflect only to a limited extent the diversity in real-world patient populations. We propose a sequential hierarchical Bayesian inference framework that enables continuous cross-patient learning to learn the underlying model parameters of the target patient population. It is important to note that the approach only requires summary information of the individual patient data to update the model. This separation of the individual inference from population inference enables implementation across different centers of care. The proposed approaches substantially improve current MIPD approaches, taking into account new trends in health care and aspects of practical applicability. They enable progress towards more informed clinical decision-making, ultimately increasing patient benefits beyond the current practice.

One third of the world's population lives in areas where earthquakes causing at least slight damage are frequently expected. Thus, the development and testing of global seismicity models is essential to improving seismic hazard estimates and earthquake-preparedness protocols for effective disaster-risk mitigation. Currently, the availability and quality of geodetic data along plate-boundary regions provides the opportunity to construct global models of plate motion and strain rate, which can be translated into global maps of forecasted seismicity. Moreover, the broad coverage of existing earthquake catalogs facilitates in present-day the calibration and testing of global seismicity models. As a result, modern global seismicity models can integrate two independent factors necessary for physics-based, long-term earthquake forecasting, namely interseismic crustal strain accumulation and sudden lithospheric stress release. In this dissertation, I present the construction of and testing results for two global ensemble seismicity models, aimed at providing mean rates of shallow (0-70 km) earthquake activity for seismic hazard assessment. These models depend on the Subduction Megathrust Earthquake Rate Forecast (SMERF2), a stationary seismicity approach for subduction zones, based on the conservation of moment principle and the use of regional "geodesy-to-seismicity" parameters, such as corner magnitudes, seismogenic thicknesses and subduction dip angles. Specifically, this interface-earthquake model combines geodetic strain rates with instrumentally-recorded seismicity to compute long-term rates of seismic and geodetic moment. Based on this, I derive analytical solutions for seismic coupling and earthquake activity, which provide this earthquake model with the initial abilities to properly forecast interface seismicity. Then, I integrate SMERF2 interface-seismicity estimates with earthquake computations in non-subduction zones provided by the Seismic Hazard Inferred From Tectonics based on the second iteration of the Global Strain Rate Map seismicity approach to construct the global Tectonic Earthquake Activity Model (TEAM). Thus, TEAM is designed to reduce number, and potentially spatial, earthquake inconsistencies of its predecessor tectonic earthquake model during the 2015-2017 period. Also, I combine this new geodetic-based earthquake approach with a global smoothed-seismicity model to create the World Hybrid Earthquake Estimates based on Likelihood scores (WHEEL) model. This updated hybrid model serves as an alternative earthquake-rate approach to the Global Earthquake Activity Rate model for forecasting long-term rates of shallow seismicity everywhere on Earth. Global seismicity models provide scientific hypotheses about when and where earthquakes may occur, and how big they might be. Nonetheless, the veracity of these hypotheses can only be either confirmed or rejected after prospective forecast evaluation. Therefore, I finally test the consistency and relative performance of these global seismicity models with independent observations recorded during the 2014-2019 pseudo-prospective evaluation period. As a result, hybrid earthquake models based on both geodesy and seismicity are the most informative seismicity models during the testing time frame, as they obtain higher information scores than their constituent model components. These results support the combination of interseismic strain measurements with earthquake-catalog data for improved seismicity modeling. However, further prospective evaluations are required to more accurately describe the capacities of these global ensemble seismicity models to forecast longer-term earthquake activity.

In today's world, many applications produce large amounts of data at an enormous rate. Analyzing such datasets for metadata is indispensable for effectively understanding, storing, querying, manipulating, and mining them. Metadata summarizes technical properties of a dataset which rang from basic statistics to complex structures describing data dependencies. One type of dependencies is inclusion dependency (IND), which expresses subset-relationships between attributes of datasets. Therefore, inclusion dependencies are important for many data management applications in terms of data integration, query optimization, schema redesign, or integrity checking. So, the discovery of inclusion dependencies in unknown or legacy datasets is at the core of any data profiling effort. For exhaustively detecting all INDs in large datasets, we developed S-indd++, a new algorithm that eliminates the shortcomings of existing IND-detection algorithms and significantly outperforms them. S-indd++ is based on a novel concept for the attribute clustering for efficiently deriving INDs. Inferring INDs from our attribute clustering eliminates all redundant operations caused by other algorithms. S-indd++ is also based on a novel partitioning strategy that enables discording a large number of candidates in early phases of the discovering process. Moreover, S-indd++ does not require to fit a partition into the main memory--this is a highly appreciable property in the face of ever-growing datasets. S-indd++ reduces up to 50% of the runtime of the state-of-the-art approach. None of the approach for discovering INDs is appropriate for the application on dynamic datasets; they can not update the INDs after an update of the dataset without reprocessing it entirely. To this end, we developed the first approach for incrementally updating INDs in frequently changing datasets. We achieved that by reducing the problem of incrementally updating INDs to the incrementally updating the attribute clustering from which all INDs are efficiently derivable. We realized the update of the clusters by designing new operations to be applied to the clusters after every data update. The incremental update of INDs reduces the time of the complete rediscovery by up to 99.999%. All existing algorithms for discovering n-ary INDs are based on the principle of candidate generation--they generate candidates and test their validity in the given data instance. The major disadvantage of this technique is the exponentially growing number of database accesses in terms of SQL queries required for validation. We devised Mind2, the first approach for discovering n-ary INDs without candidate generation. Mind2 is based on a new mathematical framework developed in this thesis for computing the maximum INDs from which all other n-ary INDs are derivable. The experiments showed that Mind2 is significantly more scalable and effective than hypergraph-based algorithms.

Landslides are frequent natural hazards in rugged terrain, when the resisting frictional force of the surface of rupture yields to the gravitational force. These forces are functions of geological and morphological factors, such as angle of internal friction, local slope gradient or curvature, which remain static over hundreds of years; whereas more dynamic triggering events, such as rainfall and earthquakes, compromise the force balance by temporarily reducing resisting forces or adding transient loads. This thesis investigates landslide distribution and orientation due to landslide triggers (e.g. rainfall) at different scales (6-4∙10^5 km^2) and aims to link rainfall movement with the landslide distribution. It additionally explores the local impacts of the extreme rainstorms on landsliding and the role of precursory stability conditions that could be induced by an earlier trigger, such as an earthquake. Extreme rainfall is a common landslide trigger. Although several studies assessed rainfall intensity and duration to study the distribution of thus triggered landslides, only a few case studies quantified spatial rainfall patterns (i.e. orographic effect). Quantifying the regional trajectories of extreme rainfall could aid predicting landslide prone regions in Japan. To this end, I combined a non-linear correlation metric, namely event synchronization, and radial statistics to assess the general pattern of extreme rainfall tracks over distances of hundreds of kilometers using satellite based rainfall estimates. Results showed that, although the increase in rainfall intensity and duration positively correlates with landslide occurrence, the trajectories of typhoons and frontal storms were insufficient to explain landslide distribution in Japan. Extreme rainfall trajectories inclined northwestwards and were concentrated along some certain locations, such as coastlines of southern Japan, which was unnoticed in the landslide distribution of about 5000 rainfall-triggered landslides. These landslides seemed to respond to the mean annual rainfall rates. Above mentioned findings suggest further investigation on a more local scale to better understand the mechanistic response of landscape to extreme rainfall in terms of landslides. On May 2016 intense rainfall struck southern Germany triggering high waters and landslides. The highest damage was reported at the Braunsbach, which is located on the tributary-mouth fan formed by the Orlacher Bach. Orlacher Bach is a ~3 km long creek that drains a catchment of about ~6 km^2. I visited this catchment in June 2016 and mapped 48 landslides along the creek. Such high landslide activity was not reported in the nearby catchments within ~3300 km^2, despite similar rainfall intensity and duration based on weather radar estimates. My hypothesis was that several landslides were triggered by rainfall-triggered flash floods that undercut hillslope toes along the Orlacher Bach. I found that morphometric features such as slope and curvature play an important role in landslide distribution on this micro scale study site (<10 km^2). In addition, the high number of landslides along the Orlacher Bach could also be boosted by accumulated damages on hillslopes due karst weathering over longer time scales. Precursory damages on hillslopes could also be induced by past triggering events that effect landscape evolution, but this interaction is hard to assess independently from the latest trigger. For example, an earthquake might influence the evolution of a landscape decades long, besides its direct impacts, such as landslides that follow the earthquake. Here I studied the consequences of the 2016 Kumamoto Earthquake (MW 7.1) that triggered some 1500 landslides in an area of ~4000 km^2 in central Kyushu, Japan. Topography, i.e. local slope and curvature, both amplified and attenuated seismic waves, thus controlling the failure mechanism of those landslides (e.g. progressive). I found that topography fails in explaining the distribution and the preferred orientation of the landslides after the earthquake; instead the landslides were concentrated around the northeast of the rupture area and faced mostly normal to the rupture plane. This preferred location of the landslides was dominated mainly by the directivity effect of the strike-slip earthquake, which is the propagation of wave energy along the fault in the rupture direction; whereas amplitude variations of the seismic radiation altered the preferred orientation. I suspect that the earthquake directivity and the asymmetry of seismic radiation damaged hillslopes at those preferred locations increasing landslide susceptibility. Hence a future weak triggering event, e.g. scattered rainfall, could further trigger landslides at those damaged hillslopes.