A5039645491- Topic Modeling
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Human-Automation Interaction and Safety
- Augmented Reality Applications
- Virtual Reality Applications and Impacts
- Thermal Radiation and Cooling Technologies
- Adversarial Robustness in Machine Learning
- Advanced Bandit Algorithms Research
- Technology Assessment and Management
- Explainable Artificial Intelligence (XAI)
- AI-based Problem Solving and Planning
- Near-Field Optical Microscopy
- Optical properties and cooling technologies in crystalline materials
- Electrowetting and Microfluidic Technologies
- Quantum Electrodynamics and Casimir Effect
- Recommender Systems and Techniques
- Microfluidic and Bio-sensing Technologies
- Domain Adaptation and Few-Shot Learning
- Plasmonic and Surface Plasmon Research
- Urban Heat Island Mitigation
- Machine Learning and Data Classification
- Speech Recognition and Synthesis
- Quantum Mechanics and Applications
- Cold Atom Physics and Bose-Einstein Condensates
University of Rostock
2001-2020
University of Tübingen
2019-2020
Fraunhofer Institute for Communication, Information Processing and Ergonomics
2019-2020
University of Oxford
2012
State-of-the-art solutions in the areas of "Language Modelling & Generating Text", "Speech Recognition", "Generating Image Descriptions" or "Video Tagging" have been using Recurrent Neural Networks as foundation for their approaches. Understanding underlying concepts is therefore tremendous importance if we want to keep up with recent upcoming publications those areas. In this work give a short overview over some most important realm which enables readers easily understand fundamentals...
Choosing the optimizer is considered to be among most crucial design decisions in deep learning, and it not an easy one. The growing literature now lists hundreds of optimization methods. In absence clear theoretical guidance conclusive empirical evidence, decision often made based on anecdotes. this work, we aim replace these anecdotes, if with a ranking, then at least evidence-backed heuristics. To do so, perform extensive, standardized benchmark fifteen particularly popular learning...
Data augmentation is an important component in the robustness evaluation of models natural language processing (NLP) and enhancing diversity data they are trained on. In this paper, we present NL-Augmenter, a new participatory Python-based framework which supports creation both transformations (modifications to data) filters (data splits according specific features). We describe initial set 117 23 for variety tasks. demonstrate efficacy NL-Augmenter by using several its analyze popular...
We have extended the classical hydrodynamics formalism to include nonlocal quantum behavior via phenomenological Bohm potential. then solved equations derive an expression for dynamical structure factor, which describes density-density correlations in system. This can be applied high density strongly coupled electron fluids long wavelength domain. show that at densities above 7×10(25) cm(-3) there are significant differences dispersion relation. Future experiments large laser facilities...
Multilingual Machine Translation promises to improve translation quality between non-English languages. This is advantageous for several reasons, namely lower latency (no need translate twice), and reduced error cascades (e.g., avoiding losing gender formality information when translating through English).On the downside, adding more languages reduces model capacity per language, which usually countered by increasing overall size, making training harder inference slower.In this work, we...
Non-autoregressive approaches aim to improve the inference speed of translation models by only requiring a single forward pass generate output sequence instead iteratively producing each predicted token. Consequently, their quality still tends be inferior autoregressive counterparts due several issues involving token interdependence. In this work, we take step back and revisit techniques that have been proposed for improving non-autoregressive compare combined implications under third-party...
We construct the dyadic Green function of electromagnetic field in presence a spatially dispersive spherical object. It contains all information on light-scattering processes off sphere, and, particular, its imaginary part gives local density states. Our construction automatically yields dispersion relation for surface polaritons terms impedances. From methodological point view, our approach is based extinction theorem and Huygens' principle. This allows straightforward application Maxwell...
Circular spatial filtering velocimetry (CSFV) was tested during the microscopic registration of individual rotations baker's yeast cells. Their frequency-dependent rotation (electrorotation; ER) induced in rotating electric fields, which were generated a glass chip chamber with four electrodes (600 μm tip-to-tip distance). The driven sinusoidal quadrature signals 5 or 8 VPP frequencies up to 3 MHz. observed cell order 1–100 s per revolution. At each measuring frequency, independent 20 cells...
Due to more and recent technologies, better simulations can be developed, which create a greater fusion between the real virtual world thus, also generate higher degree of immersion. Higher levels immersion provide meaningful realistic results faster integration assessment new concepts (e.g. design-concepts). For this, tangible VR simulator based on an armored vehicle driver's workplace has been developed. With this simulator, user interact not only with (virtual reality) but reality....
Abstract Single-shot x-ray imaging of short-lived nanostructures such as clusters and nanoparticles near a phase transition or non-crystalizing objects large proteins viruses is currently the most elegant method for characterizing their structure. Using hard radiation provides scattering images that encode two-dimensional projections, which can be combined to identify full three-dimensional object structure from multiple identical samples. Wide-angle using XUV soft x-rays, despite yielding...
We calculate numerically the heat transfer rate between a spatially dispersive sphere and half-space. By utilising Huygens' principle extinction theorem, we derive necessary reflection coefficients at plate without need to resort additional boundary conditions. find for small distances $d\sim 1$nm significant modification of spectral due spatial dispersion. As consequence, spurious divergencies that occur in local approach are absent.
Multilingual Machine Translation promises to improve translation quality between non-English languages. This is advantageous for several reasons, namely lower latency (no need translate twice), and reduced error cascades (e.g., avoiding losing gender formality information when translating through English). On the downside, adding more languages reduces model capacity per language, which usually countered by increasing overall size, making training harder inference slower. In this work, we...
Structured State Spaces for Sequences (S4) is a recently proposed sequence model with successful applications in various tasks, e.g. vision, language modeling, and audio. Thanks to its mathematical formulation, it compresses input single hidden state, able capture long range dependencies while avoiding the need an attention mechanism. In this work, we apply S4 Machine Translation (MT), evaluate several encoder-decoder variants on WMT'14 WMT'16. contrast success find that lags behind...
Non-autoregressive approaches aim to improve the inference speed of translation models by only requiring a single forward pass generate output sequence instead iteratively producing each predicted token. Consequently, their quality still tends be inferior autoregressive counterparts due several issues involving token interdependence. In this work, we take step back and revisit techniques that have been proposed for improving non-autoregressive compare combined implications under third-party...
Traditionally, for most machine learning settings, gaining some degree of explainability that tries to give users more insights into how and why the network arrives at its predictions, restricts underlying model hinders performance a certain degree. For example, decision trees are thought as being explainable than deep neural networks but they lack on visual tasks. In this work, we empirically demonstrate applying methods architectures from literature can, in fact, achieve state-of-the-art...
In their work, Dean, Rich, and Recht create a model to research recourse availability of items in recommender system. We used the definition predictive multiplicity by Marx, Pin Calmon, Ustun examine different variations this model, using values for two parameters. Pairwise comparison models show, that most these produce very similar results terms discrepancy ambiguity only some cases sets differ significantly.