A5079476030- Stochastic processes and financial applications
- Adversarial Robustness in Machine Learning
- Scientific Measurement and Uncertainty Evaluation
- Explainable Artificial Intelligence (XAI)
- Advanced Statistical Methods and Models
- Advanced Mathematical Physics Problems
- Advanced Statistical Process Monitoring
- Stochastic processes and statistical mechanics
- Gaussian Processes and Bayesian Inference
- Muon and positron interactions and applications
- Radioactive Decay and Measurement Techniques
- Advanced MEMS and NEMS Technologies
- Radiomics and Machine Learning in Medical Imaging
- Civil and Structural Engineering Research
- advanced mathematical theories
- Bayesian Methods and Mixture Models
- Advanced Sensor Technologies Research
- Anomaly Detection Techniques and Applications
- Mathematical and Theoretical Analysis
- Advanced Malware Detection Techniques
- Machine Learning and Data Classification
- Public Administration and Political Analysis
- Physics and Engineering Research Articles
- Random Matrices and Applications
- Physical Unclonable Functions (PUFs) and Hardware Security
Physikalisch-Technische Bundesanstalt
1999-2024
Humboldt-Universität zu Berlin
2019
Klinikum Lippe
2005
Max Planck Society
1989
Max Planck Institute for the Study of Crime, Security and Law
1989
Abstract A new paradigm has emerged recently in financial modeling: rough (stochastic) volatility. First observed by Gatheral et al. high‐frequency data, subsequently derived within market microstructure models, volatility captures parsimoniously key‐stylized facts of the entire implied surface, including extreme skews (as earlier Alòs al.) that were thought to be outside scope stochastic models. On mathematical side, Markovianity and, partially, semimartingality are lost. In this paper, we...
Nous développons une version discrète de la théorie des distributions paracontrôlées comme outil pour déduire les limites d'échelles modèles discrets, et nous proposons formulation dans espaces Besov avec poids. De plus, obtenons approche martingale contrôler systématiquement moments polynômes variables aléatoires i.i.d., leurs d'échelles. Comme application, un résultat d'universalité faible le modèle parabolique d'Anderson est obtenu : étudions non linéaire d'une population potentiel...
The lattice parameter of silicon plays an important role in the determination Avogadro constant and fine-structure constant. Today, three values d220 spacing are available, measured at Physikalisch-Technische Bundesanstalt (PTB, Germany), Istituto di Metrologia G. Colonnetti (IMGC, Italy) National Research Laboratory Metrology (NRLM, Japan) based on metre scale. Using PTB comparator, spacings different materials were compared with one another order to check possibility combining results form...
Abstract In deep learning-based image classification, the entropy of a neural network’s output is often taken as measure its uncertainty. We introduce an explainability method that identifies those features in input impact most this Learning corresponding by straightforward backpropagation typically leads to results are hard interpret. propose extension recently proposed oriented, modified integrated gradients (OMIG) technique alternative produce perturbations have visual quality comparable...
We show here how the methods recently applied in Debussche and Weber (2018 Electron. J. Probab. 23 28) to solve stochastic nonlinear Schrödinger equation on can be enhanced yield solutions if nonlinearity is weak enough. prove that remain localized compact time intervals which allow us apply energy full space.
The values for the Avogadro constant N/sub A/, derived from lattice spacing, density, and molar mass of silicon single crystals at several metrological institutes, show significant differences. To illuminate this discrepancy, comparison measurements spacing density were performed. Positron annihilation used to estimate vacancy concentration in Si samples. All lead nearly same after correction influence carbon oxygen impurities. It is proved that discrepancy determinations cannot be explained...
Abstract The application of deep learning has recently been proposed for the assessment image quality in mammography. It was demonstrated a proof-of-principle study that approach can be more efficient than currently applied automated conventional methods. However, contrast to methods, black-box nature and, before it recommended routine use, must understood thoroughly. For this purpose, we propose and apply new explainability method: oriented, modified integrated gradients (OMIG) method....
Abstract The absolute length of a single-crystal silicon gauge block was measured by interferometry in the temperature range between 285 K and 320 at different air pressures from atmospheric conditions down to 10 −5 hPa. From obtained dataset, coefficient thermal expansion (CTE) determined as well compressibility—or bulk modulus—of consideration systematic correction refractometer used. As choice underlying model for evaluation is not unambiguous, Bayesian averaging approach applied take...
A Bayesian treatment of deep learning allows for the computation uncertainties associated with predictions neural networks. We show how concept Errors-in-Variables can be used in regression to also account uncertainty input employed network. The presented approach thereby exploits a relevant, but generally overlooked, source and yields decomposition predictive into an aleatoric epistemic part that is more complete and, many cases, consistent from statistical perspective. discuss along...
Abstract Evaluating a neural network on an input that differs markedly from the training data might cause erratic and flawed predictions. We study method judges unusualness of by evaluating its informative content compared to learned parameters. This technique can be used judge whether is suitable for processing certain raise red flag unexpected behavior lie ahead. compare our approach various methods uncertainty evaluation literature datasets scenarios. Specifically, we introduce simple,...
Abstract We propose a framework for the assessment of uncertainty quantification in deep regression. The is based on regression problems where function linear combination nonlinear functions. Basically, any level complexity can be realized through choice functions and dimensionality their domain. Results an are compared against those obtained by statistical reference method. method utilizes knowledge about underlying Bayesian using prior reference. flexibility, together with availability...
Model uncertainty can become a critical issue in the presence of several plausible models. Due to its definition by derivative coefficient thermal expansion is quite vulnerable when it comes model choice. Using length measurements single-crystal silicon temperature range we study results physical and polynomial model. While have consistency fits observe noticeable difference for derived between these
Abstract We consider Bayesian sample size determination using a criterion that utilizes the first two moments of posterior variance. study resulting in dependence on chosen prior and explore success rate for bounding variance below prescribed limit under true sampling distribution. Compared with based average proposed leads to an increase significantly improved rates. Generic asymptotic properties are proven, such as expression sort phase transition. Our is illustrated real world datasets...