Showing papers by "Johannes Kepler University of Linz published in 2016"

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)

Abstract: We introduce the "exponential linear unit" (ELU) which speeds up learning in deep neural networks and leads to higher classification accuracies. Like rectified linear units (ReLUs), leaky ReLUs (LReLUs) and parametrized ReLUs (PReLUs), ELUs alleviate the vanishing gradient problem via the identity for positive values. However, ELUs have improved learning characteristics compared to the units with other activation functions. In contrast to ReLUs, ELUs have negative values which allows them to push mean unit activations closer to zero like batch normalization but with lower computational complexity. Mean shifts toward zero speed up learning by bringing the normal gradient closer to the unit natural gradient because of a reduced bias shift effect. While LReLUs and PReLUs have negative values, too, they do not ensure a noise-robust deactivation state. ELUs saturate to a negative value with smaller inputs and thereby decrease the forward propagated variation and information. Therefore, ELUs code the degree of presence of particular phenomena in the input, while they do not quantitatively model the degree of their absence. In experiments, ELUs lead not only to faster learning, but also to significantly better generalization performance than ReLUs and LReLUs on networks with more than 5 layers. On CIFAR-100 ELUs networks significantly outperform ReLU networks with batch normalization while batch normalization does not improve ELU networks. ELU networks are among the top 10 reported CIFAR-10 results and yield the best published result on CIFAR-100, without resorting to multi-view evaluation or model averaging. On ImageNet, ELU networks considerably speed up learning compared to a ReLU network with the same architecture, obtaining less than 10% classification error for a single crop, single model network.

DeepTox: Toxicity Prediction using Deep Learning

Abstract: The Tox21 Data Challenge has been the largest effort of the scientific community to compare computational methods for toxicity prediction. This challenge comprised 12,000 environmental chemicals and drugs which were measured for 12 different toxic effects by specifically designed assays. We participated in this challenge to assess the performance of Deep Learning in computational toxicity prediction. Deep Learning has already revolutionized image processing, speech recognition, and language understanding but has not yet been applied to computational toxicity. Deep Learning is founded on novel algorithms and architectures for artificial neural networks together with the recent availability of very fast computers and massive datasets. It discovers multiple levels of distributed representations of the input, with higher levels representing more abstract concepts. We hypothesized that the construction of a hierarchy of chemical features gives Deep Learning the edge over other toxicity prediction methods. Furthermore, Deep Learning naturally enables multi-task learning, that is, learning of all toxic effects in one neural network and thereby learning of highly informative chemical features. In order to utilize Deep Learning for toxicity prediction, we have developed the DeepTox pipeline. First, DeepTox normalizes the chemical representations of the compounds. Then it computes a large number of chemical descriptors that are used as input to machine learning methods. In its next step, DeepTox trains models, evaluates them, and combines the best of them to ensembles. Finally, DeepTox predicts the toxicity of new compounds. In the Tox21 Data Challenge, DeepTox had the highest performance of all computational methods winning the grand challenge, the nuclear receptor panel, the stress response panel, and six single assays (teams ``Bioinf@JKU''). We found that Deep Learning excelled in toxicity prediction and outperformed many other computational approaches like naive Bayes, support vector machines, and random forests.

A Leadless Intracardiac Transcatheter Pacing System

Abstract: BackgroundA leadless intracardiac transcatheter pacing system has been designed to avoid the need for a pacemaker pocket and transvenous lead. MethodsIn a prospective multicenter study without controls, a transcatheter pacemaker was implanted in patients who had guideline-based indications for ventricular pacing. The analysis of the primary end points began when 300 patients reached 6 months of follow-up. The primary safety end point was freedom from system-related or procedure-related major complications. The primary efficacy end point was the percentage of patients with low and stable pacing capture thresholds at 6 months (≤2.0 V at a pulse width of 0.24 msec and an increase of ≤1.5 V from the time of implantation). The safety and efficacy end points were evaluated against performance goals (based on historical data) of 83% and 80%, respectively. We also performed a post hoc analysis in which the rates of major complications were compared with those in a control cohort of 2667 patients with transvenous ...

Water Determines the Structure and Dynamics of Proteins

Abstract: Water is an essential participant in the stability, structure, dynamics, and function of proteins and other biomolecules. Thermodynamically, changes in the aqueous environment affect the stability of biomolecules. Structurally, water participates chemically in the catalytic function of proteins and nucleic acids and physically in the collapse of the protein chain during folding through hydrophobic collapse and mediates binding through the hydrogen bond in complex formation. Water is a partner that slaves the dynamics of proteins, and water interaction with proteins affect their dynamics. Here we provide a review of the experimental and computational advances over the past decade in understanding the role of water in the dynamics, structure, and function of proteins. We focus on the combination of X-ray and neutron crystallography, NMR, terahertz spectroscopy, mass spectroscopy, thermodynamics, and computer simulations to reveal how water assist proteins in their function. The recent advances in computer s...

Business Models for Sustainability A Co-Evolutionary Analysis of Sustainable Entrepreneurship, Innovation, and Transformation

Abstract: The relevance of business models for corporate performance in general and corporate sustainability in particular has been widely acknowledged in the literature while sustainable entrepreneurship re...

A Randomized Trial of Factor VIII and Neutralizing Antibodies in Hemophilia A

Abstract: BackgroundThe development of neutralizing anti–factor VIII alloantibodies (inhibitors) in patients with severe hemophilia A may depend on the concentrate used for replacement therapy. MethodsWe conducted a randomized trial to assess the incidence of factor VIII inhibitors among patients treated with plasma-derived factor VIII containing von Willebrand factor or recombinant factor VIII. Patients who met the eligibility criteria (male sex, age <6 years, severe hemophilia A, and no previous treatment with any factor VIII concentrate or only minimal treatment with blood components) were included from 42 sites. ResultsOf 303 patients screened, 264 underwent randomization and 251 were analyzed. Inhibitors developed in 76 patients, 50 of whom had high-titer inhibitors (≥5 Bethesda units). Inhibitors developed in 29 of the 125 patients treated with plasma-derived factor VIII (20 patients had high-titer inhibitors) and in 47 of the 126 patients treated with recombinant factor VIII (30 patients had high-titer inhib...

Aging-related tau astrogliopathy (ARTAG): harmonized evaluation strategy

Abstract: Pathological accumulation of abnormally phosphorylated tau protein in astrocytes is a frequent, but poorly characterized feature of the aging brain. Its etiology is uncertain, but its presence is sufficiently ubiquitous to merit further characterization and classification, which may stimulate clinicopathological studies and research into its pathobiology. This paper aims to harmonize evaluation and nomenclature of aging-related tau astrogliopathy (ARTAG), a term that refers to a morphological spectrum of astroglial pathology detected by tau immunohistochemistry, especially with phosphorylation-dependent and 4R isoform-specific antibodies. ARTAG occurs mainly, but not exclusively, in individuals over 60 years of age. Tau-immunoreactive astrocytes in ARTAG include thorn-shaped astrocytes at the glia limitans and in white matter, as well as solitary or clustered astrocytes with perinuclear cytoplasmic tau immunoreactivity that extends into the astroglial processes as fine fibrillar or granular immunopositivity, typically in gray matter. Various forms of ARTAG may coexist in the same brain and might reflect different pathogenic processes. Based on morphology and anatomical distribution, ARTAG can be distinguished from primary tauopathies, but may be concurrent with primary tauopathies or other disorders. We recommend four steps for evaluation of ARTAG: (1) identification of five types based on the location of either morphologies of tau astrogliopathy: subpial, subependymal, perivascular, white matter, gray matter; (2) documentation of the regional involvement: medial temporal lobe, lobar (frontal, parietal, occipital, lateral temporal), subcortical, brainstem; (3) documentation of the severity of tau astrogliopathy; and (4) description of subregional involvement. Some types of ARTAG may underlie neurological symptoms; however, the clinical significance of ARTAG is currently uncertain and awaits further studies. The goal of this proposal is to raise awareness of astroglial tau pathology in the aged brain, facilitating communication among neuropathologists and researchers, and informing interpretation of clinical biomarkers and imaging studies that focus on tau-related indicators.

Protons and Hydroxide Ions in Aqueous Systems

Abstract: Understanding the structure and dynamics of water's constituent ions, proton and hydroxide, has been a subject of numerous experimental and theoretical studies over the last century. Besides their obvious importance in acid-base chemistry, these ions play an important role in numerous applications ranging from enzyme catalysis to environmental chemistry. Despite a long history of research, many fundamental issues regarding their properties continue to be an active area of research. Here, we provide a review of the experimental and theoretical advances made in the last several decades in understanding the structure, dynamics, and transport of the proton and hydroxide ions in different aqueous environments, ranging from water clusters to the bulk liquid and its interfaces with hydrophobic surfaces. The propensity of these ions to accumulate at hydrophobic surfaces has been a subject of intense debate, and we highlight the open issues and challenges in this area. Biological applications reviewed include proton transport along the hydration layer of various membranes and through channel proteins, problems that are at the core of cellular bioenergetics.

Multiple-stable anisotropic magnetoresistance memory in antiferromagnetic MnTe

Abstract: Commercial magnetic memories rely on the bistability of ordered spins in ferromagnetic materials. Recently, experimental bistable memories have been realized using fully compensated antiferromagnetic metals. Here we demonstrate a multiple-stable memory device in epitaxial MnTe, an antiferromagnetic counterpart of common II–VI semiconductors. Favourable micromagnetic characteristics of MnTe allow us to demonstrate a smoothly varying zero-field antiferromagnetic anisotropic magnetoresistance (AMR) with a harmonic angular dependence on the writing magnetic field angle, analogous to ferromagnets. The continuously varying AMR provides means for the electrical read-out of multiple-stable antiferromagnetic memory states, which we set by heat-assisted magneto-recording and by changing the writing field direction. The multiple stability in our memory is ascribed to different distributions of domains with the Neel vector aligned along one of the three magnetic easy axes. The robustness against strong magnetic field perturbations combined with the multiple stability of the magnetic memory states are unique properties of antiferromagnets. Contrary to ferromagnets, antiferromagnets possess no net magnetic moment, which has limited their applicability as magnetic memory media. Here, the authors demonstrate a heat-assisted multiple-stable memory based on epitaxial thin films of antiferromagnet MnTe with three-fold symmetric anisotropy.

Preparation of polyphosphazenes: a tutorial review

Abstract: Poly(organo)phosphazenes are a family of inorganic molecular hybrid polymers with very diverse properties due to the vast array of organic substituents possible. This tutorial review aims to introduce the basics of the synthetic chemistry of polyphosphazenes, detailing for readers outside the field the essential knowledge required to design and prepare polyphosphazenes with desired properties. A particular focus is given to some of the recent advances in their chemical synthesis which allows not only the preparation of polyphosphazenes with controlled molecular weights and polydispersities, but also novel branched architectures and block copolymers. We also discuss the preparation of supramolecular structures, bioconjugates and in situ forming gels from this diverse family of functional materials. This tutorial review aims to equip the reader to prepare defined polyphosphazenes with unique property combinations and in doing so we hope to stimulate further research and yet more innovative applications for these highly interesting multifaceted materials.

On the Efficiency Limit of Conjugated Polymer:Fullerene-Based Bulk Heterojunction Solar Cells.

Abstract: The power conversion efficiency potential of eight high-performance polymer-fullerene blends is investigated. All studied absorbers show the typical organic solar cell losses limiting their performance to ≈13%.

Immigration and Voting for the Far Right

Abstract: Does the presence of immigrants in one's neighbourhood affect voting for far right-wing parties? We study the case of the Freedom Party of Austria (FPO) which, under the leadership of Jorg Haider, increased its vote share from less than 5 percent in the early 1980s to 27 percent by the end of the 1990s and continued to attract more than 20 percent of voters in the 2013 national election. We find that the inflow of immigrants into a community has a significant impact on the increase in the community's voting share for the FPO, explaining roughly a tenth of the regional variation in vote changes. Our results suggest that voters worry about adverse labor market effects of immigration, as well as about the quality of their neighbourhood. In fact, we find evidence of a negative impact of immigration on "compositional amenities." In communities with larger immigration influx, Austrian children commute longer distances to school, and fewer daycare resources are provided. We do not find evidence that Austrians move out of communities with increasing immigrant presence.

Ultraflexible organic amplifier with biocompatible gel electrodes

Abstract: In vivo electronic monitoring systems are promising technology to obtain biosignals with high spatiotemporal resolution and sensitivity. Here we demonstrate the fabrication of a biocompatible highly conductive gel composite comprising multi-walled carbon nanotube-dispersed sheet with an aqueous hydrogel. This gel composite exhibits admittance of 100 mS cm(-2) and maintains high admittance even in a low-frequency range. On implantation into a living hypodermal tissue for 4 weeks, it showed a small foreign-body reaction compared with widely used metal electrodes. Capitalizing on the multi-functional gel composite, we fabricated an ultrathin and mechanically flexible organic active matrix amplifier on a 1.2-μm-thick polyethylene-naphthalate film to amplify (amplification factor: ∼200) weak biosignals. The composite was integrated to the amplifier to realize a direct lead epicardial electrocardiography that is easily spread over an uneven heart tissue.

The LFM-1b Dataset for Music Retrieval and Recommendation

Abstract: We present the LFM-1b dataset of more than one billion music listening events created by more than 120,000 users of Last.fm. Each listening event is characterized by artist, album, and track name, and further includes a timestamp. On the (anonymous) user level, basic demographics and a selection of more elaborate user descriptors are included. The dataset is foremost intended for benchmarking in music information retrieval and recommendation. To facilitate experimentation in a straightforward manner, it also includes a precomputed user-item-playcount matrix. In addition, sample Python scripts showing how to load the data and perform efficient computations are provided. An implementation of a simple collaborative filtering recommender rounds off the code package. We discuss in detail the LFM-1b dataset's acquisition, availability, statistics, and content, and place it in the context of existing datasets. We also showcase its usage in a simple artist recommendation task, whose results are intended to serve as baseline against which more elaborate techniques can be assessed. The two unique features of the dataset in comparison to existing ones are (i) its substantial size and (ii) a wide range of additional user descriptors that reflect their music taste and consumption behavior.

madmom: A New Python Audio and Music Signal Processing Library

Abstract: In this paper, we present madmom, an open-source audio processing and music information retrieval (MIR) library written in Python. madmom features a concise, NumPy-compatible, object oriented design with simple calling conventions and sensible default values for all parameters, which facilitates fast prototyping of MIR applications. Prototypes can be seamlessly converted into callable processing pipelines through madmom's concept of Processors, callable objects that run transparently on multiple cores. Processors can also be serialised, saved, and re-run to allow results to be easily reproduced anywhere. Apart from low-level audio processing, madmom puts emphasis on musically meaningful high-level features. Many of these incorporate machine learning techniques and madmom provides a module that implements some methods commonly used in MIR such as hidden Markov models and neural networks. Additionally, madmom comes with several state-of-the-art MIR algorithms for onset detection, beat, downbeat and meter tracking, tempo estimation, and chord recognition. These can easily be incorporated into bigger MIR systems or run as stand-alone programs.

Wavelength-tunable sources of entangled photons interfaced with atomic vapours.

Abstract: The prospect of using the quantum nature of light for secure communication keeps spurring the search and investigation of suitable sources of entangled photons. A single semiconductor quantum dot is one of the most attractive, as it can generate indistinguishable entangled photons deterministically and is compatible with current photonic-integration technologies. However, the lack of control over the energy of the entangled photons is hampering the exploitation of dissimilar quantum dots in protocols requiring the teleportation of quantum entanglement over remote locations. Here we introduce quantum dot-based sources of polarization-entangled photons whose energy can be tuned via three-directional strain engineering without degrading the degree of entanglement of the photon pairs. As a test-bench for quantum communication, we interface quantum dots with clouds of atomic vapours, and we demonstrate slow-entangled photons from a single quantum emitter. These results pave the way towards the implementation of hybrid quantum networks where entanglement is distributed among distant parties using optoelectronic devices.

madmom: a new Python Audio and Music Signal Processing Library

Abstract: In this paper, we present madmom, an open-source audio processing and music information retrieval (MIR) library written in Python. madmom features a concise, NumPy-compatible, object oriented design with simple calling conventions and sensible default values for all parameters, which facilitates fast prototyping of MIR applications. Prototypes can be seamlessly converted into callable processing pipelines through madmom's concept of Processors, callable objects that run transparently on multiple cores. Processors can also be serialised, saved, and re-run to allow results to be easily reproduced anywhere. Apart from low-level audio processing, madmom puts emphasis on musically meaningful high-level features. Many of these incorporate machine learning techniques and madmom provides a module that implements some in MIR commonly used methods such as hidden Markov models and neural networks. Additionally, madmom comes with several state-of-the-art MIR algorithms for onset detection, beat, downbeat and meter tracking, tempo estimation, and piano transcription. These can easily be incorporated into bigger MIR systems or run as stand-alone programs.

Enhancing clone-and-own with systematic reuse for developing software variants.

Abstract: To keep pace with the increasing demand for custom-tailored software systems, companies often apply a practice called clone-and-own, whereby a new variant of a software system is built by coping and adapting existing variants. Instead of a single and configurable system, clone-and-own leads to ad hoc product portfolios of multiple yet similar variants that soon become impossible to maintain effectively. Clone-and-own has widespread industrial use because it requires no major upfront investments and is intuitive, but it lacks a methodology for systematic reuse. In this work we propose ECCO (Extraction and Composition for Clone-and-Own), a novel approach to enhance clone and-own that actively supports the development and maintenance of software product variants. A software engineer selects the desired features and ECCO finds the proper software artifacts to reuse and then provides guidance during the manual completion by hinting which software artifacts may need adaptation. We evaluated our approach on 6 case studies, covering 402 variants having up to 344KLOC, and found that precision and recall of composed products quickly reach a near optimum (>95% reuse).

Fusing Social Media Cues: Personality Prediction from Twitter and Instagram

Abstract: Incorporating users' personality traits has shown to be instrumental in many personalized retrieval and recommender systems. Analysis of users' digital traces has become an important resource for inferring personality traits. To date, the analysis of users' explicit and latent characteristics is typically restricted to a single social networking site (SNS). In this work, we propose a novel method that integrates text, image, and users' meta features from two different SNSs: Twitter and Instagram. Our preliminary results indicate that the joint analysis of users' simultaneous activities in two popular SNSs seems to lead to a consistent decrease of the prediction errors for each personality trait.

Is it sex or personality? the impact of sex stereotypes on discrimination in applicant selection

Abstract: This paper goes one step further than previous experimental studies by adopting correspondence testing to investigate the impact of usually unobservable variables. When testing for discrimination it may not be sufficient to control for human capital only. Specific personality traits that are more commonly associated with men than women in general seem to contribute to success particularly in many attractive, highly paid jobs. A successful manager, for example, is supposed to be ambitious, competitive, and dominant, which are stereotypically masculine traits. Alternatively, stereotypically feminine characteristics are preferred in many traditionally female occupations. A good nurse or kindergarten teacher, for example, seems to require

Thermally stable coexistence of liquid and solid phases in gallium nanoparticles

Abstract: A real-time investigation shows that Ga nanoparticles in the solid γ-phase coexist with liquid Ga at a broad range of temperatures, as a result of nanoscale confinement, Laplace pressure and epitaxial matching with the substrate.

Nonmagnetic band gap at the Dirac point of the magnetic topological insulator (Bi1−xMnx)2Se3

Abstract: Magnetic doping is expected to open a band gap at the Dirac point of topological insulators by breaking time-reversal symmetry and to enable novel topological phases. Epitaxial (Bi1-xMnx)(2)Se-3 is a prototypical magnetic topological insulator with a pronounced surface band gap of similar to 100 meV. We show that this gap is neither due to ferromagnetic order in the bulk or at the surface nor to the local magnetic moment of the Mn, making the system unsuitable for realizing the novel phases. We further show that Mn doping does not affect the inverted bulk band gap and the system remains topologically nontrivial. We suggest that strong resonant scattering processes cause the gap at the Dirac point and support this by the observation of in-gap states using resonant photoemission. Our findings establish a mechanism for gap opening in topological surface states which challenges the currently known conditions for topological protection.

Molecular mechanisms of STIM/Orai communication.

Abstract: Ca2+ entry into the cell via store-operated Ca2+ release-activated Ca2+ (CRAC) channels triggers diverse signaling cascades that affect cellular processes like cell growth, gene regulation, secreti...