A5038355567- Adversarial Robustness in Machine Learning
- Advanced Neural Network Applications
- Lung Cancer Treatments and Mutations
- Cancer Genomics and Diagnostics
- Anomaly Detection Techniques and Applications
- Dark Matter and Cosmic Phenomena
- Astrophysics and Cosmic Phenomena
- Neural Networks and Applications
- Neutrino Physics Research
- Advanced Image Processing Techniques
- Pancreatic and Hepatic Oncology Research
- CCD and CMOS Imaging Sensors
- Hip and Femur Fractures
- Machine Learning and Data Classification
- Computational Physics and Python Applications
- Emergency and Acute Care Studies
- Cancer Immunotherapy and Biomarkers
- Radioactivity and Radon Measurements
- Advanced Data Storage Technologies
- Scientific Computing and Data Management
- Machine Learning and Algorithms
- Trauma and Emergency Care Studies
- Domain Adaptation and Few-Shot Learning
- COVID-19 diagnosis using AI
- Generative Adversarial Networks and Image Synthesis
Universitätsklinikum Gießen und Marburg
2022
RWTH Aachen University
2018-2019
Optimization of Binarized Neural Networks (BNNs) currently relies on real-valued latent weights to accumulate small update steps. In this paper, we argue that these cannot be treated analogously in networks. Instead their main role is provide inertia during training. We interpret current methods terms and novel insights into the optimization BNNs. subsequently introduce first optimizer specifically designed for BNNs, Binary Optimizer (Bop), demonstrate its performance CIFAR-10 ImageNet....
We introduce Larq Compute Engine, the world's fastest Binarized Neural Network (BNN) inference engine, and use this framework to investigate several important questions about efficiency of BNNs design a new state-of-the-art BNN architecture. LCE provides highly optimized implementations binary operations accelerates convolutions by 8.5 - 18.5x compared their full-precision counterparts on Pixel 1 phones. LCE's integration with sophisticated MLIR-based converter allow users move smoothly from...
The German Society for Geriatrics recommends the "ISAR" questionnaire as a screening tool patients ≥70 geriatric in emergency rooms. Although ISAR-score is collected routinely "AltersTraumaRegister DGU®" (ATR-DGU), to date less known about predictive value of "ISAR"-score trauma patients.Currently, 84 clinics participate ATR-DGU. This evaluation limited subgroup proximal femur fractures from 2016-2018. Patients years, who underwent surgery hip fracture are included In this evaluation,...
We present a novel approach to search for origins of ultra-high energy cosmic rays. These particles are likely nuclei that initiate extensive air showers in the Earth's atmosphere. In large-area observatories, particle arrival directions measured together with their energies and atmospheric depth at which maximize. The depths provide rough measures nuclear charges. simultaneous fit all observed rays we use galactic magnetic field as mass spectrometer adapt charges such extragalactic...
VISPA (Visual Physics Analysis) is a web-platform that enables users to work on any secure shell (SSH) reachable resource using just their webbrowser. It used successfully in research and education for HEP data analysis. The emerging JupyterLab an ideal choice comprehensive, browser-based, extensible environment we seek unify it with the efforts of VISPA-project. primary objective provide user freedom access external resources at disposal, while maintaining smooth integration preconfigured...
We use adversarial network architectures together with the Wasserstein distance to generate or refine simulated detector data. The data reflect two-dimensional projections of spatially distributed signal patterns a broad spectrum applications. As an example, we observatory detect cosmic ray-induced air showers ground-based array particle detectors. First investigate method generating variable strengths while constraining primary energy. then present technique time traces detectors match...
We present a novel approach to search for origins of ultra-high energy cosmic rays. In simultaneous fit all observed rays we use the galactic magnetic field as mass spectrometer and adapt nuclear charges such that their extragalactic arrival directions are concentrated in few possible. During constraint by individual shower depth measurements. show simulated astrophysical scenario source can be reconstructed even within substantial isotropic background.