Showing papers by "Brno University of Technology published in 2012"
Posted Content•
TL;DR: This paper proposes a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem and validates empirically the hypothesis and proposed solutions.
Abstract: There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in Bengio et al. (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed solutions in the experimental section.
3,549 citations
TL;DR: It is shown that on the basis of open-source software development, a fully functional software package can be created that covers the needs of a large part of the scanning probe microscopy user community.
Abstract: In this article, we review special features of Gwyddion—a modular, multiplatform, open-source software for scanning probe microscopy data processing, which is available at http://gwyddion.net/. We describe its architecture with emphasis on modularity and easy integration of the provided algorithms into other software. Special functionalities, such as data processing from non-rectangular areas, grain and particle analysis, and metrology support are discussed as well. It is shown that on the basis of open-source software development, a fully functional software package can be created that covers the needs of a large part of the scanning probe microscopy user community.
3,151 citations
01 Dec 2012
TL;DR: This paper improves recurrent neural network language models performance by providing a contextual real-valued input vector in association with each word to convey contextual information about the sentence being modeled by performing Latent Dirichlet Allocation using a block of preceding text.
Abstract: Recurrent neural network language models (RNNLMs) have recently demonstrated state-of-the-art performance across a variety of tasks. In this paper, we improve their performance by providing a contextual real-valued input vector in association with each word. This vector is used to convey contextual information about the sentence being modeled. By performing Latent Dirichlet Allocation using a block of preceding text, we achieve a topic-conditioned RNNLM. This approach has the key advantage of avoiding the data fragmentation associated with building multiple topic models on different data subsets. We report perplexity results on the Penn Treebank data, where we achieve a new state-of-the-art. We further apply the model to the Wall Street Journal speech recognition task, where we observe improvements in word-error-rate.
644 citations
09 Jul 2012
TL;DR: It is shown, on this mobile phone database, that face and speaker recognition can be performed in a mobile environment and using score fusion can improve the performance by more than 25% in terms of error rates.
Abstract: This paper presents a novel fully automatic bi-modal, face and speaker, recognition system which runs in real-time on a mobile phone. The implemented system runs in real-time on a Nokia N900 and demonstrates the feasibility of performing both automatic face and speaker recognition on a mobile phone. We evaluate this recognition system on a novel publicly-available mobile phone database and provide a well defined evaluation protocol. This database was captured almost exclusively using mobile phones and aims to improve research into deploying biometric techniques to mobile devices. We show, on this mobile phone database, that face and speaker recognition can be performed in a mobile environment and using score fusion can improve the performance by more than 25% in terms of error rates.
235 citations
01 Dec 2012
TL;DR: This paper presents novel language-independent bottleneck (BN) feature extraction framework, where each language is modelled by separate output layer, while all the hidden layers jointly model the variability of all the source languages.
Abstract: In this paper we present novel language-independent bottleneck (BN) feature extraction framework. In our experiments we have used Multilingual Artificial Neural Network (ANN), where each language is modelled by separate output layer, while all the hidden layers jointly model the variability of all the source languages. The key idea is that the entire ANN is trained on all the languages simultaneously, thus the BN-features are not biased towards any of the languages. Exactly for this reason, the final BN-features are considered as language independent. In the experiments with GlobalPhone database, we show that Multilingual BN-features consistently outperform Monolingual BN-features. Also, cross-lingual generalization is evaluated, where we train on 5 source languages and test on 3 other languages. The results show that the ANN can produce very good BN-features even for unseen languages, in some cases even better than if we trained the ANN on the target language only.
217 citations
TL;DR: In this review, the connection between zinc(II) ions, reactive oxygen species, heavy metal ions and metallothioneins is demonstrated with respect to effect of these proteins on cell proliferation and a possible negative role in resistance to heavy metal-based and non-heavy metal- based drugs.
Abstract: Metallothioneins (MT) are a family of ubiquitous proteins, whose role is still discussed in numerous papers, but their affinity to some metal ions is undisputable These cysteine-rich proteins are connected with antioxidant activity and protective effects on biomolecules against free radicals, especially reactive oxygen species In this review, the connection between zinc(II) ions, reactive oxygen species, heavy metal ions and metallothioneins is demonstrated with respect to effect of these proteins on cell proliferation and a possible negative role in resistance to heavy metal-based and non-heavy metal-based drugs
213 citations
TL;DR: Using ANN for automatic sleep scoring is especially promising because of new ANN learning algorithms allowing faster classification without decreasing the performance.
197 citations
TL;DR: In this article, the thermal insulation from sheep wool has been tested under various conditions and the building physics and acoustic properties were specifically tested which are important for durable and undamaged applications.
167 citations
TL;DR: The use of laser-induced breakdown spectroscopy (LIBS) for trace element determination in different matrices is reviewed in this article, where the main emphasis is on spatially resolved analysis of microbiological, plant and animal samples.
155 citations
TL;DR: In this paper, an overview of thermal treatment methods for waste-to-energy (WTE) processes technologies in terms of their performance and environmental impact is presented. But the authors focus on the potential of waste treatments and related legislation by the European Communities.
147 citations
TL;DR: An overview of the AMIDA systems for transcription of conference and lecture room meetings, developed for participation in the Rich Transcription evaluations conducted by the National Institute for Standards and Technology in the years 2007 and 2009 is given.
Abstract: In this paper, we give an overview of the AMIDA systems for transcription of conference and lecture room meetings. The systems were developed for participation in the Rich Transcription evaluations conducted by the National Institute for Standards and Technology in the years 2007 and 2009 and can process close talking and far field microphone recordings. The paper first discusses fundamental properties of meeting data with special focus on the AMI/AMIDA corpora. This is followed by a description and analysis of improved processing and modeling, with focus on techniques specifically addressing meeting transcription issues such as multi-room recordings or domain variability. In 2007 and 2009, two different strategies of systems building were followed. While in 2007 we used our traditional style system design based on cross adaptation, the 2009 systems were constructed semi-automatically, supported by improved decoders and a new method for system representation. Overall these changes gave a 6%-13% relative reduction in word error rate compared to our 2007 results while at the same time requiring less training material and reducing the real-time factor by five times. The meeting transcription systems are available at www.webasr.org.
TL;DR: In this paper, the basic physical characteristics, mechanical and fracture-mechanics properties, durability characteristics, hydric and thermal properties of high performance concrete (HPC) with up to 60% of Portland cement replaced by fine-ground ceramics.
Abstract: This paper presents experimental work regarding the basic physical characteristics, mechanical and fracture-mechanics properties, durability characteristics, hydric and thermal properties of high performance concrete (HPC) with up to 60% of Portland cement replaced by fine-ground ceramics. Experimental results show that the amount of the ceramics in the mix is limited mainly by the resistance against de-icing salts which is found satisfactory only up to the cement replacement level of 10%. The mechanical and water transport properties are not significantly impaired by ceramic additions of up to 20%, whereas the effective fracture toughness, specific fracture energy, and chemical resistance (to MgCl2 ,N H 4Cl, Na2SO4, HCl) are effectively maintained up to 40%. The frost resistance, water vapor transport and storage parameters and thermal properties are not significantly impaired even up to a 60% replacement level. 2011 Elsevier Ltd. All rights reserved.
25 Mar 2012
TL;DR: A lattice generation method that is exact, i.e. it satisfies all the natural properties the authors would want from a lattice of alternative transcriptions of an utterance, and does not introduce substantial overhead above one-best decoding.
Abstract: We describe a lattice generation method that is exact, i.e. it satisfies all the natural properties we would want from a lattice of alternative transcriptions of an utterance. This method does not introduce substantial overhead above one-best decoding. Our method is most directly applicable when using WFST decoders where the WFST is “fully expanded”, i.e. where the arcs correspond to HMM transitions. It outputs lattices that include HMM-state-level alignments as well as word labels. The general idea is to create a state-level lattice during decoding, and to do a special form of determinization that retains only the best-scoring path for each word sequence. This special determinization algorithm is a solution to the following problem: Given a WFST A, compute a WFST B that, for each input-symbol-sequence of A, contains just the lowest-cost path through A.
Proceedings Article•
01 Jan 2012TL;DR: It is shown that significant gains in SAD accuracy can be obtained by careful design of acoustic front end, feature normalization, incorporation of long span features via data-driven dimensionality reducing transforms, and channel dependent modeling.
Abstract: This paper describes the speech activity detection (SAD) system developed by the Patrol team for the first phase of the DARPA RATS (Robust Automatic Transcription of Speech) program, which seeks to advance state of the art detection capabilities on audio from highly degraded communication channels. We present two approaches to SAD, one based on Gaussian mixture models, and one based on multi-layer perceptrons. We show that significant gains in SAD accuracy can be obtained by careful design of acoustic front end, feature normalization, incorporation of long span features via data-driven dimensionality reducing transforms, and channel dependent modeling. We also present a novel technique for normalizing detection scores from different systems for the purpose of system combination.
28 Mar 2012
TL;DR: Inspired by machine learning approaches in biometric person authentication, an offline framework for task-independent prediction of interaction intents is developed and tested and the principles of the method, the features extracted, normalization methods, and evaluation metrics are described.
Abstract: Interaction intent prediction and the Midas touch have been a longstanding challenge for eye-tracking researchers and users of gaze-based interaction. Inspired by machine learning approaches in biometric person authentication, we developed and tested an offline framework for task-independent prediction of interaction intents. We describe the principles of the method, the features extracted, normalization methods, and evaluation metrics. We systematically evaluated the proposed approach on an example dataset of gaze-augmented problem-solving sessions. We present results of three normalization methods, different feature sets and fusion of multiple feature types. Our results show that accuracy of up to 76% can be achieved with Area Under Curve around 80%. We discuss the possibility of applying the results for an online system capable of interaction intent prediction.
TL;DR: In this article, an HPLC-ICPMS method based on sample extraction with trifluoroacetic acid/H 2 O 2, and measurement of arsenate by anion-exchange HPLC and ICPMS using aqueous malonic acid as mobile phase was presented.
TL;DR: Red yeast strains cultivated at optimal growth conditions and in medium with modified carbon and nitrogen sources produced dried carotenoid-enriched red yeast biomass which could be directly used in feed industry as nutrition supplement.
TL;DR: Yeasts and yeast-like organisms associated with matured fruits and fully open blossoms of apple, plum, and pear trees, during 2 consecutive years at 3 localities in southwest Slovakia were focused on.
Abstract: Yeasts are common inhabitants of the phyllosphere, but our knowledge of their diversity in various plant organs is still limited. This study focused on the diversity of yeasts and yeast-like organi...
TL;DR: An extensive set of parameters of several lime-metakaolin plasters, including basic material characteristics, mechanical and fracture-mechanical properties, durability characteristics, hydric parameters and thermal properties, is presented in this paper.
TL;DR: In this article, it was shown that defects such as Eu${}^{3+}$ and oxygen vacancies strongly influence the temperature of the phase transition to antiferrodistortive phase as well as the tendency to incommensurate modulation in EuTiO${}_{3}$ ceramics.
Abstract: X-ray diffraction, dynamical mechanical analysis, and infrared reflectivity studies revealed an antiferrodistortive phase transition in EuTiO${}_{3}$ ceramics. Near 300 K, the perovskite structure changes from cubic $Pm\overline{3}m$ to tetragonal $I4/mcm$ due to antiphase tilting of oxygen octahedra along the $\mathbf{c}$ axis (${a}^{0}{a}^{0}{c}^{\ensuremath{-}}$ in Glazer notation). The phase transition is analogous to SrTiO${}_{3}$. However, some ceramics as well as single crystals of EuTiO${}_{3}$ show different infrared reflectivity spectra bringing evidence of a different crystal structure. In such samples, electron diffraction revealed an incommensurate tetragonal structure with modulation wave vector q $\ensuremath{\simeq}$ 0.38 a${}^{*}$. Extra phonons in samples with modulated structure are activated in the IR spectra due to folding of the Brillouin zone. We propose that defects such as Eu${}^{3+}$ and oxygen vacancies strongly influence the temperature of the phase transition to antiferrodistortive phase as well as the tendency to incommensurate modulation in EuTiO${}_{3}$.
TL;DR: The article provides deeper and more accurate analysis than can be found in the literature, including the memory complexity, on the generalization of the Goertzel algorithm, which allows it to be used for frequencies which are not integer multiples of the fundamental frequency.
Abstract: The article deals with the Goertzel algorithm, used to establish the modulus and phase of harmonic components of a signal. The advantages of the Goertzel approach over the DFT and the FFT in cases of a few harmonics of interest are highlighted, with the article providing deeper and more accurate analysis than can be found in the literature, including the memory complexity. But the main emphasis is placed on the generalization of the Goertzel algorithm, which allows us to use it also for frequencies which are not integer multiples of the fundamental frequency. Such an algorithm is derived at the cost of negligibly increasing the computational and memory complexity.
TL;DR: The paper establishes several results concerning jumping finite automata in terms of commonly investigated areas of automata theory, such as decidability and closure properties, and achieves several results that demonstrate differences between jumping finiteAutomata and classical finite Automata.
Abstract: The present paper proposes a new investigation area in automata theory — jumping finite automata. These automata work like classical finite automata except that they read input words discontinuously — that is, after reading a symbol, they can jump over some symbols within the words and continue their computation from there. The paper establishes several results concerning jumping finite automata in terms of commonly investigated areas of automata theory, such as decidability and closure properties. Most importantly, it achieves several results that demonstrate differences between jumping finite automata and classical finite automata. In its conclusion, the paper formulates several open problems and suggests future investigation areas.
27 Aug 2012
TL;DR: A publicly available toolkit and a benchmark suite for rigorous verification of Integer Numerical Transition Systems (INTS), which can be viewed as control-flow graphs whose edges are annotated by Presburger arithmetic formulas.
Abstract: This paper presents a publicly available toolkit and a benchmark suite for rigorous verification of Integer Numerical Transition Systems (INTS), which can be viewed as control-flow graphs whose edges are annotated by Presburger arithmetic formulas. We present Flata and Eldarica, two verification tools for INTS. The Flata system is based on precise acceleration of the transition relation, while the Eldarica system is based on predicate abstraction with interpolation-based counterexample-driven refinement. The Eldarica verifier uses the Princess theorem prover as a sound and complete interpolating prover for Presburger arithmetic. Both systems can solve several examples for which previous approaches failed, and present a useful baseline for verifying integer programs. The infrastructure is a starting point for rigorous benchmarking, competitions, and standardized communication between tools.
TL;DR: In this article, the early microstructural changes leading to fatigue crack initiation in cyclically strained polycrystals (nickel, 316L steel) were investigated in detail using electron channeling contrast imaging (ECCI) technique (concurrently in the FIB crosssection and on the specimen surface) and simultaneously with the surface relief topography using transmission electron microscopy (TEM) of thin surface foils prepared by in situ lift-out technique.
TL;DR: In this article, it was shown that the system of three difference equations, where all elements of the sequences were real numbers, can be solved, and some consequences on asymptotic behavior of solutions for the case when coefficients are periodic with period three were deduced.
Abstract: We show that the system of three difference equations , , and , , where all elements of the sequences , , , , , and initial values , , , , are real numbers, can be solved. Explicit formulae for solutions of the system are derived, and some consequences on asymptotic behavior of solutions for the case when coefficients are periodic with period three are deduced.
25 Jun 2012
TL;DR: Two techniques of normalization based on total, between- and within-speaker variance spectra 1 normalize the i-vectors length for Gaussianity, but the first adapts the ivectors representation to a speaker recognition system based on LDA and two-covariance scoring when the second adapts it to a Gaussian-PLDA model.
Abstract: I-vector extraction and Probabilistic Linear Discriminant Analysis (PLDA) has become the state-of-the-art configuration for speaker verification. Recently, Gaussian-PLDA has been improved by a preliminary length normalization of i-vectors. This normalization, known to increase the Gaussianity of the i-vector distribution, also improves performance of systems based on standard Linear Discriminant Analysis (LDA) and ”two-covariance model” scoring. But this technique follows a standardization of the i-vectors (centering and whitening ivectors based on the first and second order moments of the development data). We propose in this paper two techniques of normalization based on total, between- and within-speaker variance spectra 1 . These ”spectral” techniques both normalize the i-vectors length for Gaussianity, but the first adapts the ivectors representation to a speaker recognition system based on LDA and two-covariance scoring when the second adapts it to a Gaussian-PLDA model. Significant performance improvements are demonstrated on the male and female telephone portion of NIST SRE 2010. Index Terms: i-vectors, probabilistic linear discriminant analysis, speaker recognition.
01 Apr 2012
TL;DR: In this paper, the authors presented new topologies for realizing one lossless grounded inductor and two floating inductors employing a single differential difference current conveyor and a minimum number of passive components, two resistors, and one grounded capacitor.
Abstract: In this work, we present new topologies for realizing one lossless grounded inductor and two floating, one lossless and one lossy, inductors employing a single differential difference current conveyor (DDCC) and a minimum number of passive components, two resistors, and one grounded capacitor. The floating inductors are based on ordinary dual-output differential difference cur- rent conveyor (DO-DDCC) while the grounded lossless inductor is based one a modified dual-output differential difference current conveyor (MDO-DDCC). The proposed lossless floating inductor is obtained from the lossy one by employing a negative impedance converter (NIC). The non-ideality effects of the active element on the simulated inductors are investigated. To demonstrate the perform- ance of the proposed grounded inductance simulator as an example, it is used to construct a parallel resonant circuit. SPICE simulation results are given to confirm the theoretical analysis.
25 Mar 2012
TL;DR: The purpose of this task was to perform audio search with audio input in four languages, with very few resources being available in each language.
Abstract: In this paper, we describe the “Spoken Web Search” Task, which was held as part of the 2011 MediaEval benchmark campaign. The purpose of this task was to perform audio search with audio input in four languages, with very few resources being available in each language. The data was taken from “spoken web” material collected over mobile phone connections by IBM India. We present results from several independent systems, developed by five teams and using different approaches, compare them, and provide analysis and directions for future research.
TL;DR: The intended aim of this work is to create a database for simple and fast identification of archeological or paleontological materials in situ, which can speed up and simplify the sampling process during archeological excavations that nowadays tend to be quite damaging and timeconsuming.
TL;DR: A novel finding is published in this brief, namely, that the area within the loop is directly related to the value of action potential, which was introduced by Leon Chua in his original work from 1971.
Abstract: It is well known that the memristor driven by a periodical voltage or current exhibits pinched v - i hysteresis loop. A novel finding is published in this brief, namely, that the area within the loop is directly related to the value of action potential, which was introduced by Leon Chua in his original work from 1971.