Villmann T.,Mittweida University of Applied Sciences |
Haase S.,Mittweida University of Applied Sciences
Neural Computation | Year: 2011
Supervised and unsupervised vector quantization methods for classification and clustering traditionally use dissimilarities, frequently taken as Euclidean distances. In this article, we investigate the applicability of divergences instead, focusing on online learning. We deduce the mathematical fundamentals for its utilization in gradient-based online vector quantization algorithms. It bears on the generalized derivatives of the divergences known as Fréchet derivatives in functional analysis, which reduces in finite-dimensional problems to partial derivatives in a natural way.We demonstrate the application of this methodology for widely applied supervised and unsupervised online vector quantization schemes, including self-organizing maps, neural gas, and learning vector quantization. Additionally, principles for hyperparameter optimization and relevance learning for parameterized divergences in the case of supervised vector quantization are given to achieve improved classification accuracy. © 2011 Massachusetts Institute of Technology.
Biehl M.,University of Groningen |
Hammer B.,Bielefeld University |
Villmann T.,Mittweida University of Applied Sciences
Wiley Interdisciplinary Reviews: Cognitive Science | Year: 2016
An overview is given of prototype-based models in machine learning. In this framework, observations, i.e., data, are stored in terms of typical representatives. Together with a suitable measure of similarity, the systems can be employed in the context of unsupervised and supervised analysis of potentially high-dimensional, complex datasets. We discuss basic schemes of competitive vector quantization as well as the so-called neural gas approach and Kohonen's topology-preserving self-organizing map. Supervised learning in prototype systems is exemplified in terms of learning vector quantization. Most frequently, the familiar Euclidean distance serves as a dissimilarity measure. We present extensions of the framework to nonstandard measures and give an introduction to the use of adaptive distances in relevance learning. © 2016 Wiley Periodicals, Inc.
Agency: European Commission | Branch: FP7 | Program: CP | Phase: ENERGY.2012.10.2.1 | Award Amount: 3.91M | Year: 2012
CyanoFactory brings together ten selected leading, highly complementary European partners with the aim to carry out integrated, fundamental research aiming at applying synthetic biology principles towards a cell factory notion in microbial biotechnology. The vision is to build on recent progress in synthetic biology and develop novel photosynthetic cyanobacteria as chassis to be used as self-sustained cell factories in generating a solar fuel. This will include the development of a toolbox with orthogonal parts and devices for cyanobacterial synthetic biology, improvement of the chassis enabling enhanced growth and robustness in challenging environmental conditions, establishment of a data warehouse facilitating the modelling and optimization of cyanobacterial metabolic pathways, and strong and novel bioinformatics for effective data mining. To reach the goal, a combination of basic and applied R&D is needed; basic research to design and construct the cyanobacterial cells efficiently evolving H2 from the endless resources solar energy and water, and applied research to design and construct the advanced photobioreactors that efficiently produce H2. Biosafety is of highest concern and dedicated efforts will be made to address and control cell survival and death. The aim, to develop a (photo)synthetic cell factory, will have an enormous impact on the future options and possibilities for renewable solar fuel production. The consortium includes academic, research institute and industry participants with the direct involvement of two SMEs in the advanced photobioreactor design, construction and use. Purpose-designed, specifically engineered self-sustained cells utilising solar energy and CO2 from the air, may be the mechanisms and processes by which we generate large scale renewable energy carriers in our future societies. CyanoFactory offers Europe the possibility to take a lead, and not only follow, in these very important future and emerging technologies!
Agency: European Commission | Branch: H2020 | Program: ECSEL-IA | Phase: ECSEL-14-2015 | Award Amount: 61.99M | Year: 2016
Addressing European Policies for 2020 and beyond the Power Semiconductor and Electronics Manufacturing 4.0 (SemI40) project responds to the urgent need of increasing the competitiveness of the Semiconductor manufacturing industry in Europe through establishing smart, sustainable, and integrated ECS manufacturing. SemI40 will further pave the way for serving highly innovative electronic markets with products powered by microelectronics Made in Europe. Positioned as an Innovation Action it is the high ambition of SemI40 to implement technical solutions on TRL level 4-8 into the pilot lines of the industry partners. Challenging use cases will be implemented in real manufacturing environment considering also their technical, social and economic impact to the society, future working conditions and skills needed. Applying Industry 4.0, Big Data, and Industrial Internet technologies in the electronics field requires holistic and complex actions. The selected main objectives of SemI40 covered by the MASP2015 are: balancing system security and production flexibility, increase information transparency between fields and enterprise resource planning (ERP), manage critical knowledge for improved decision making and maintenance, improve fab digitalization and virtualization, and enable automation systems for agile distributed production. SemI40s value chain oriented consortium consists of 37 project partners from 5 European countries. SemI40 involves a vertical and horizontal supply chain and spans expertise and partners from raw material research, process and assembly innovation and pilot line, up to various application domains representing enhanced smart systems. Through advancing manufacturing of electronic components and systems, SemI40 contributes to safeguard more than 20.000 jobs of people directly employed in the participating facilities, and in total more than 300.000 jobs of people employed at all industry partners facilities worldwide.
Schneider K.A.,Mittweida University of Applied Sciences |
Schneider K.A.,University of Vienna |
Escalante A.A.,Arizona State University
Malaria Journal | Year: 2013
Background: Considering the distinct biological characteristics of Plasmodium species is crucial for control and elimination efforts, in particular when facing the spread of drug resistance. Whereas the evolutionary fitness of all malarial species could be approximated by the probability of being taken by a mosquito and then infecting a new host, the actual steps in the malaria life cycle leading to a successful transmission event show differences among Plasmodium species. These "steps" are called fitness components. Differences in terms of fitness components may affect how selection imposed by interventions, e.g. drug treatments, differentially acts on each Plasmodium species. Thus, a successful malaria control or elimination programme should understand how differences in fitness components among different malaria species could affect adaptive evolution (e.g. the emergence of drug resistance). In this investigation, the interactions between some fitness components and natural selection are explored. Methods. A population-genetic model is formulated that qualitatively explains how different fitness components (in particular gametocytogenesis and longevity of gametocytes) affect selection acting on merozoites during the erythrocytic cycle. By comparing Plasmodium falciparum and Plasmodium vivax, the interplay of parasitaemia and gametocytaemia dynamics in determining fitness is modelled under circumstances that allow contrasting solely the differences between these two parasites in terms of their fitness components. Results: By simulating fitness components, it is shown that selection acting on merozoites (e.g., on drug resistant mutations or malaria antigens) is more efficient in P. falciparum than in P. vivax. These results could explain, at least in part, why resistance against drugs, such as chloroquine (CQ) is highly prevalent in P. falciparum worldwide, while CQ is still a successful treatment for P. vivax despite its massive use. Furthermore, these analyses are used to explore the importance of understanding the dynamic of gametocytaemia to ascertain the spreading of drug resistance. Conclusions: The strength of natural selection on mutations that express their advantage at the merozoite stage is different in P. vivax and P. falciparum. Species-specific differences in gametocytogenesis and longevity of gametocytes need to be accounted for when designing effective malaria control and elimination programmes. There is a need for reliable data on gametocytogenesis from field studies. © 2013 Schneider and Escalante; licensee BioMed Central Ltd.
Schwobbermeyer H.,SunGene GmbH |
Wunschiers R.,Mittweida University of Applied Sciences
Methods in Molecular Biology | Year: 2012
Data from high-throughput experimental methods are currently being used to construct complex biological networks. These include regulatory gene networks, regulatory protein-DNA networks, protein-protein interaction networks, or metabolic networks. Independent of its type, every network can be characterized by a number of parameters such as number of nodes, number of edges connecting nodes, direction and weight of edges, in- and out-degree of nodes, etc. One can draw an analogy of such rather simple network parameters to the primary sequence of proteins or nucleic acids. More insight can be gained by an analysis of the secondary and tertiary structure of biomolecules, which often contain motifs. The same holds for biological networks. The occurrence and frequency of certain motifs or pattern characterize the topology and often the functional space of a network. Here, we describe the utilization of the free software MAVisto, which was designed to mine networks for typical motifs by combining a flexible motif search algorithm with interactive exploration methods and sophisticated visualization techniques. © 2012 Springer Science+Business Media, LLC.
Foulquier J.,Mittweida University of Applied Sciences
12th Conference-Seminar: International School on Nonsinusoidal Currents and Compensation, ISNCC 2015 - Conference Proceedings | Year: 2015
This paper presents the inverter control of a shunt active power filter (APF) for the mitigation of harmonics caused by multiple low power Harmonic Generating Loads (HGLs) in a Low Voltage (LV) grid. The shunt APF acts as adjustable admittance for each individual harmonic frequency in accordance with the harmonic voltage and the grid admittance at its installation point. The aim of the compensation is to maintain the harmonic voltages under the standard threshold and to collect the harmonic currents from the HGLs installed close to the shunt APF by shaping an optimal compensation current. The harmonic detection method and the controller strategy are described and experimental result on a prototype attests the performances of the harmonics compensation. © 2015 IEEE.
Kastner M.,Mittweida University of Applied Sciences |
Villmann T.,Mittweida University of Applied Sciences
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2012
In this paper we propose a new approach to combine unsupervised and supervised vector quantization for clustering and fuzzy classification using the framework of neural vector quantizers like self-organizing maps or neural gas. For this purpose the original cost functions are modified in such a way that both aspects, unsupervised vector quantization and supervised classification, are incorporated. The theoretical justification of the convergence of the new algorithm is given by an adequate redefinition of the underlying dissimilarity measure now interpreted as a dissimilarity in the data space combined with the class label space. This allows a gradient descent learning as known for the original algorithms. Thus a semi-supervised learning scheme is achieved. We apply this method for a spectra image cube of remote sensing data for landtype classification. The obtained fuzzy class visualizations allow a better understanding and interpretation of the spectra. © 2012 Springer-Verlag Berlin Heidelberg.
Heinke F.,Mittweida University of Applied Sciences |
Labudde D.,Mittweida University of Applied Sciences
Computational and Mathematical Methods in Medicine | Year: 2012
Diabetes insipidus (DI) is a rare endocrine, inheritable disorder with low incidences in an estimated one per 25,000-30,000 live births. This disease is characterized by polyuria and compensatory polydypsia. The diverse underlying causes of DI can be central defects, in which no functional arginine vasopressin (AVP) is released from the pituitary or can be a result of defects in the kidney (nephrogenic DI, NDI). NDI is a disorder in which patients are unable to concentrate their urine despite the presence of AVP. This antidiuretic hormone regulates the process of water reabsorption from the prourine that is formed in the kidney. It binds to its type-2 receptor (V2R) in the kidney induces a cAMP-driven cascade, which leads to the insertion of aquaporin-2 water channels into the apical membrane. Mutations in the genes of V2R and aquaporin-2 often lead to NDI. We investigated a structure model of V2R in its bound and unbound state regarding protein stability using a novel protein energy profile approach. Furthermore, these techniques were applied to the wild-type and selected mutations of aquaporin-2. We show that our results correspond well to experimental water ux analysis, which confirms the applicability of our theoretical approach to equivalent problems. Copyright © 2012 Florian Heinke and Dirk Labudde.
Mittweida University of Applied Sciences | Date: 2012-02-09
The invention relates to methods and devices for breaking up ore. The methods and devices are characterised in particular in that ore mineral or ore minerals can be subsequently easily extracted. For this purpose coherent NIR radiation, non-coherent NIR radiation, at least one electric alternating field having a frequency greater than 300 GHz, at least one magnetic alternating field having a frequency greater than 300 GHz, at least one electromagnetic alternating field having a frequency greater than 300 GHz, or a combination thereof are respectively applied to the ore at least once by means of a device for generating the radiation, the at least one alternating field, or the radiation and the at least one alternating field, wherein ore mineral, ore minerals, absorbent components, or ore minerals and absorbent components of the ore absorb(s) energy from the radiation, the alternating field, or the radiation and the alternating field and said energy is not or is only slightly absorbed by the lode matter. Thus, advantageously, cracks are formed in the ore or the ore splits by means of the resulting stresses.