A5080777994- Privacy-Preserving Technologies in Data
- Neural Networks and Applications
- Stochastic Gradient Optimization Techniques
- Advanced Neural Network Applications
- Machine Learning and Algorithms
- Data Stream Mining Techniques
- Advanced Bandit Algorithms Research
- Domain Adaptation and Few-Shot Learning
- Data Visualization and Analytics
- Computational Drug Discovery Methods
- Distributed Sensor Networks and Detection Algorithms
- Age of Information Optimization
- Adversarial Robustness in Machine Learning
- Network Security and Intrusion Detection
- Explainable Artificial Intelligence (XAI)
- Machine Learning and Data Classification
- Software System Performance and Reliability
- Data Quality and Management
- Advanced Malware Detection Techniques
- AI in cancer detection
- Advanced Graph Neural Networks
- Parallel Computing and Optimization Techniques
- Machine Learning in Bioinformatics
- Cybercrime and Law Enforcement Studies
- Stock Market Forecasting Methods
Ruhr University Bochum
2022-2025
Monash University
2019-2024
Essen University Hospital
2023-2024
Institut für Medizinische Informatik, Biometrie und Epidemiologie
2022
Helmholtz Center for Information Security
2021
University of Bonn
2014-2019
Fraunhofer Institute for Intelligent Analysis and Information Systems
2013-2019
University of Pisa
2019
IBM Research - Ireland
2019
Laboratoire Traitement et Communication de l’Information
2019
In our clinical study, 200 total knee arthroplasties were evaluated to compare the use of OrthoPilot system with conventional mechanical instrumentation. Long-term outcome replacement depends mainly on accuracy implant positioning and restoration leg axis. Our experience was that navigation could achieve a greater degree concerning aforementioned aspects. Among 513 primary-inserted replacements, 100 navigated knees compared conventionally implanted after matching two groups patients by...
In many critical applications, sensitive data is inherently distributed and cannot be centralized due to privacy concerns. A wide range of federated learning approaches have been proposed train models locally at each client without sharing their data, typically by exchanging model parameters, or probabilistic predictions (soft labels) on a public dataset combination both. However, these methods still disclose private information restrict local those that can trained using gradient-based...
Identifying informative components in binary data is an essential task many application areas, including life sciences, social and recommendation systems. Boolean matrix factorization (BMF) a family of methods that performs this by factorizing the into dense factor matrices. In real-world settings, often distributed across stakeholders required to stay private, prohibiting straightforward BMF. To adapt BMF context, we approach problem from federated-learning perspective, building on...
Arguably the most representative application of artificial intelligence, autonomous driving systems usually rely on computer vision techniques to detect situations external environment. Object detection underpins ability scene understanding in such systems. However, existing object algorithms often behave as a black box, so when model fails, no information is available When, Where and How failure happened. In this paper, we propose visual analytics approach help developers interpret...
Recent advancements in Graph Neural Networks (GNNs) show promise for various applications like social networks and financial networks. However, they exhibit fairness issues, particularly human-related decision contexts, risking unfair treatment of groups historically subject to discrimination. While several visual analytics studies have explored machine learning (ML), few tackled the particular challenges posed by GNNs. We propose a framework GNN analysis, offering insights into how...
The emerging paradigm of federated learning (FL) strives to enable collaborative training deep models on the network edge without centrally aggregating raw data and hence improving privacy. In most cases, assumption independent identically distributed samples across local clients does not hold for setups. Under this setting, neural performance may vary significantly according distribution even hurt convergence. Most previous work has focused a difference in labels or client shifts. Unlike...
Prior to the deep learning era, shape was commonly used describe objects. Nowadays, state-of-the-art (SOTA) algorithms in medical imaging are predominantly diverging from computer vision, where voxel grids, meshes, point clouds, and implicit surface models used. This is seen numerous shape-related publications premier vision conferences as well growing popularity of ShapeNet (about 51,300 models) Princeton ModelNet (127,915 models). For domain, we present a large collection anatomical shapes...
Federated learning allows multiple parties to collaboratively train a joint model without sharing local data. This enables applications of machine in settings inherently distributed, undisclosable data such as the medical domain. In practice, training is usually achieved by aggregating models, for which objectives have be expectation similar (global) objective. Often, however, datasets are so small that differ greatly from global objective, resulting federated fail. We propose novel approach...
Gradient boosting of prediction rules is an efficient approach to learn potentially interpretable yet accurate probabilistic models. However, actual interpretability requires limit the number and size generated rules, existing variants are not designed for this purpose. Though corrective refits all rule weights in each iteration minimise risk, included conditions tend be sub-optimal, because commonly used objective functions fail anticipate refitting. Here, we address issue by a new function...
A Turbomachinery Gridding System (TGS) is presented that allows for setting up grids steady and unsteady multistage turbomachinery simulations. There are several unique aspects of this gridding system: 1.) it starts with an axisymmetric view the process, 2.) command driven batch oriented so new geometries can be automatically gridded, 3.) interactive component used to visualize geometries, 4.) customized real applications a minimum input required, 5.) each blade row gridded separately either...
We present a novel parallelisation scheme that simplifies the adaptation of learning algorithms to growing amounts data as well needs for accurate and confident predictions in critical applications. In contrast other techniques, it can be applied broad class without further mathematical derivations writing dedicated code, while at same time maintaining theoretical performance guarantees. Moreover, our is able reduce runtime many polylogarithmic on quasi-polynomially processing units. This...
Flatness of the loss curve is conjectured to be connected generalization ability machine learning models, in particular neural networks. While it has been empirically observed that flatness measures consistently correlate strongly with generalization, still an open theoretical problem why and under which circumstances light reparameterizations change certain but leave unchanged. We investigate connection between by relating interpolation from representative data, deriving notions...
Given multiple datasets over a fixed set of random variables, each collected from different environment, we are interested in discovering the shared underlying causal network and local interventions per without assuming prior knowledge on which observational or interventional, shape dependencies. We formalize this problem using Algorithmic Model Causation, instantiate consistent score via Minimum Description Length principle, show under conditions identifiable. To efficiently discover...
Frequent subgraph mining, i.e., the identification of relevant patterns in graph databases, is a well-known data mining problem with high practical relevance, since next to summarizing data, resulting can also be used define powerful domain-specific similarity functions for prediction. In recent years, significant progress has been made towards algorithms that scale complex graphs by focusing on tree and probabilistically allowing small amount incompleteness result. Nonetheless, complexity...
Corporate earnings are a crucial indicator for investment and business valuation. Despite their importance the fact that classic econometric approaches fail to match analyst forecasts by orders of magnitude, automatic prediction corporate from public data is not in focus current machine learning research. In this paper, we present first time fully automatized method at same a) only relies on publicly available b) can outperform human analysts. The latter shown empirically an experiment...
Corporate earnings are a crucial indicator for investment and business valuation. Despite their importance the fact that classic econometric approaches fail to match analyst forecasts by orders of magnitude, automatic prediction corporate from public data is not in focus current machine learning research. In this paper, we present first time fully automatized method at same a) only relies on publicly available b) can outperform human analysts. The latter shown empirically an experiment...