Trustworthy AI is dedicated to the development of methodologies and proofs that demonstrate the “proper” behavior of Artificial Intelligence algorithms, in order to favor their acceptance by users and organizations. By considering explainable AI and Uncertainty Quantification, we will show that defining consistent inferential procedures give systematic, reliable and arguable information for...
Machine learning has transformed many industries, being employed not only on large centralized datasets, but increasingly on data generated by a multitude of networked, complex devices such as mobile phones, autonomous vehicles or industrial machines. However, data-privacy and security concerns often prevent the centralization of this data, most prominently in healthcare. Federated learning...
Although a large amount of data is collected on each patient during cancer care, clinical decisions are mostly based on limited parameters and expert knowledge. This is mainly due to insufficient data infrastructure and a lack of tools to comprehensively integrate diverse clinical data. At University Hospital Essen, medical data is stored in FHIR format, enabling cutting-edge analyses of...
We explore the integration of panoptic scene graphs in the field of chest radiographs, to enhance explainable medical report generation. Panoptic scene graphs require a model to generate a more comprehensive scene graph representation based on panoptic segmentations rather than rigid bounding boxes and thus present a more holistic image representation. These graphs facilitate accurate report...
The use of machine learning methods in clinical settings is increasing. One reason for this is the availability of more complex models that promise more accurate predictive performance, especially for the study of heterogeneous diseases with multimodal data, such as Alzheimer’s disease. However, as machine learning models become more complex, their interpretability decreases. The reduced...
Control charts are a well-known approach for quality improvement and anomaly detection. They are applied to quality-related processes (e.g., metrics of product quality from a monitored manufacturing process) and allow to detect "deviations from normality", i.e., if the process turns from its specified in-control state into an out-of-control state. In this study, we focus on ordinal data...
An important task in reliability studies is the lifetime testing of systems consisting of dependent or interacting components. Since the fatigue of a composite material is largely determined by the failure of its components, its risk of breakdown can be linked to the observable component failure times. These failure times form a simple point process that has numerous applications also in...
When monitoring complex manufacturing processes, various methods, for instance the optimization of observed systems or quantification of their uncertainty, are applied to support and improve the processes. These methods necessitate repeated evaluations of the systems associated responses. While complex numerical models such as finite element models are capable of this, their solutions come...
In metrology, the science of measurement, as well in industrial applications in which measurement accuracy is of importance, it is required to evaluate the uncertainty of each measurement. For complex instruments like an industrial work horse as the coordinate measurement machine (CMM), evaluating the uncertainty can be a similarly complex task. To this purpose a simulation model, often...
We present an automated script, which controls the meter readings for electricity, water, heat and cold at TU Dortmund University, Germany. The script combines historic and current consumption data and calculates individual forecasts for every meter. These one-step-ahead forecasts are compared with true values afterwards to identify deviation from the regular energy consumption or anomalies....
Artificial intelligence plays an important role today. It facilitates or completely takes over office tasks such as writing, formatting, and correcting texts, or in the medical field, enabling early detection and diagnosis of diseases.
However, data provided by algorithms can significantly disadvantage certain individuals. The results of such discrimination are often noticed later and can...
The utilisation of Artificial Intelligence (AI) in medical practice is on the rise due to various factors. Firstly, they can process large datasets and recognise complex relationships that may be difficult for humans to discern in the enormous amount of medical data. Therefore AI systems can enhance the efficiency and accuracy of medical processes, thus saving resources.
Nevertheless, the...
In classical statistical process monitoring (SPM) applications the Phase I sample is assumed to come from an in-control process, which is however not always valid, especially when the monitoring characteristic for each item/case is a vector of profiles, i.e., a multivariate profile.
The presence of untrustable observations, or, in general, of outliers, especially in high-dimensional...
Quantifying the similarity between datasets has widespread applications in statistics and machine learning. The performance of a predictive model on novel datasets, referred to as generalizability, depends on how similar the training and evaluation datasets are. Exploiting or transferring insights between similar datasets is a key aspect of meta-learning and transfer-learning. In simulation...
The increasing popularity of machine learning in many application fields has led to an increasing demand in methods of explainable machine learning as they are e.g. provided by the R packages DALEX (Biecek, 2018) and iml (Molnar, 2018). A general process to ensure the development of transparent and auditable machine learning models in industry (TAX4CS) is given in Bücker et al. (2021).
In...
Advanced statistical and machine learning models methods are becoming increasingly important in applied data science. At the same time, their trustworthiness is critical for the progress and adoption of data science applications in various fields, including official statistics.
„Bad quality reduces trust very, very fast.“ Taking up this dictum, official statistics in Germany have...
AI is about to revolutionize many sectors of society and will soon have significant economic, legal, social and regulatory consequences.
In the world of production, transport, human resource management, and health, to name but a few, a growing share of diagnostic and planning processes is operated by AI-based systems.
Controlling the risks of deploying these AI for high risk systems...
Label noise, the mislabeling of instances in a dataset, is harmful to classifier performance, increases model complexity, and impairs adequate feature selection. It is frequent in large scale datasets and naturally occurs when human experts are involved. While extensive research has focused on mitigating label noise in image and text datasets through deep neural networks, there exists a...
Traditionally, ordinal response data have been modeled through parametric models such as the proportional odds model. More recently, popular machine learning methods such as random forest (RF) have been extended for ordinal prediction. As RF does not inherently support ordinal response data, a common approach is assigning numeric scores to the ordinal response categories and learning a...
Being able to quantify the importance of random inputs of an input-output black-box model is at the cornerstone of the fields of sensitivity analysis (SA) and explainable artificial intelligence (XAI). To perform this task, methods such as Shapley effects and SHAP have received a lot of attention. The former offers a solution for output variance decomposition with non-independent inputs, and...
Explainable Artificial Intelligence (XAI) stands as a crucial area of research essential for advancing AI applications in real-world contexts. Within XAI, Global Sensitivity Analysis (GSA) methods assume significance, offering insights into the influential impact of individual or grouped parameters on the predictions of machine learning models, as well as the outcomes of simulators and...
Despite attractive theoretical guarantees and practical successes, Predictive Interval (PI) given by Conformal Prediction (CP) may not reflect the uncertainty of a given model. This limitation arises from CP methods using a constant correction for all test points, disregarding their individual epistemic uncertainties, to ensure coverage properties. To address this issue, we propose using a...
Machine learning models are often the basis of current automated systems. Trust in an automated system is typically justified only up to a certain degree: A moderately reliable system deserves less trust than a highly reliable one. Ideally, trust is calibrated, in the sense that a human interacting with a system neither over- nor undertrusts the system. To be able to relate objective measures...
Machine learning (ML) will play an increasingly important role in many processes of insurance companies in the future [1]. However, ML models are at risk of being attacked and manipulated [2]. In this work, the robustness of Gradient Boosted Decision Tree (GBDT) models and Deep Neural Networks (DNN) in an insurance context is evaluated. It is analyzed how vulnerable each model is against...
When creating multi-channel time-series datasets for Human Activity Recognition (HAR), researchers are faced with the issue of subject selection criteria. It is unknown what physical characteristics and/or soft biometrics, such as age, height, and weight, must be considered to train a classifier to achieve robustness toward heterogeneous populations in the training and testing data. This...
Automated guided vehicles (AGVs) are an essential
area of research for the industry to enable dynamic transport
operations. Furthermore, AGV-based multi-robot systems (MRS)
are being utilized in various applications, e.g. in production or in
logistics. Most research today focuses on ensuring that the system
is operational, which is not always achieved. In daily use, faults
and failures...
In the evolving landscape of machine learning research, theconcept of trustworthiness receives critical consideration, both concern-ing data and models. However, the lack of a universally agreed upondefinition of the very concept of trustworthiness presents a considerablechallenge. The lack of such a definition impedes meaningful exchange andcomparison of results when it comes to assessing...
Past few years have witnessed significant leaps in capabilities of Large Language Models (LLMs). LLMs of today can perform a variety of tasks such as summarization, information retrieval and even mathematical reasoning with impressive accuracy. What is even more impressive is LLMs’ ability to follow natural language instructions without needing dedicated training datasets. However, issues like...