Reconeixement de veu mitjançant xarxes neuronals

La interacció home-màquina per mitjà de la veu cobreix moltes àrees d’investigació. Es destaquen entre altres, el reconeixement de la parla, la síntesis i identificació de discurs, la verificació i identificació de locutor i l’activació per veu (ordres) de sistemes robòtics. Reconèixer la parla és natural i simple per a les persones, però és un treball complex per a les màquines, pel qual existeixen diverses metodologies i tècniques, entre elles les Xarxes Neuronals. L’objectiu d’aquest treball és desenvolupar una eina en Matlab per al reconeixement i identificació de paraules pronunciades per un locutor, entre un conjunt de paraules possibles, i amb una bona fiabilitat dins d’uns marges preestablerts. El sistema és independent del locutor que pronuncia la paraula, és a dir, aquest locutor no haurà intervingut en el procés d’entrenament del sistema. S’ha dissenyat una interfície que permet l’adquisició del senyal de veu i el seu processament mitjançant xarxes neuronals i altres tècniques. Adaptant una part de control al sistema, es podria utilitzar per donar ordres a un robot com l’Alfa6Uvic o qualsevol altre dispositiu.

Prediction of the bulking phenomenon in wastewater treatment plants

The control and prediction of wastewater treatment plants poses an important goal: to avoid breaking the environmental balance by always keeping the system in stable operating conditions. It is known that qualitative information — coming from microscopic examinations and subjective remarks — has a deep influence on the activated sludge process. In particular, on the total amount of effluent suspended solids, one of the measures of overall plant performance. The search for an input–output model of this variable and the prediction of sudden increases (bulking episodes) is thus a central concern to ensure the fulfillment of current discharge limitations. Unfortunately, the strong interrelationbetween variables, their heterogeneity and the very high amount of missing information makes the use of traditional techniques difficult, or even impossible. Through the combined use of several methods — rough set theory and artificial neural networks, mainly — reasonable prediction models are found, which also serve to show the different importance of variables and provide insight into the process dynamics

The prediction of speed and incline in outdoor running in humans using accelerometry.

PURPOSE: To explore whether triaxial accelerometric measurements can be utilized to accurately assess speed and incline of running in free-living conditions. METHODS: Body accelerations during running were recorded at the lower back and at the heel by a portable data logger in 20 human subjects, 10 men, and 10 women. After parameterizing body accelerations, two neural networks were designed to recognize each running pattern and calculate speed and incline. Each subject ran 18 times on outdoor roads at various speeds and inclines; 12 runs were used to calibrate the neural networks whereas the 6 other runs were used to validate the model. RESULTS: A small difference between the estimated and the actual values was observed: the square root of the mean square error (RMSE) was 0.12 m x s(-1) for speed and 0.014 radiant (rad) (or 1.4% in absolute value) for incline. Multiple regression analysis allowed accurate prediction of speed (RMSE = 0.14 m x s(-1)) but not of incline (RMSE = 0.026 rad or 2.6% slope). CONCLUSION: Triaxial accelerometric measurements allows an accurate estimation of speed of running and incline of terrain (the latter with more uncertainty). This will permit the validation of the energetic results generated on the treadmill as applied to more physiological unconstrained running conditions.

Nonlinear inverse dynamic models of gas sensing systems based on chemical sensor arrays for quantitative measurements

Gas sensing systems based on low-cost chemical sensor arrays are gaining interest for the analysis of multicomponent gas mixtures. These sensors show different problems, e.g., nonlinearities and slow time-response, which can be partially solved by digital signal processing. Our approach is based on building a nonlinear inverse dynamic system. Results for different identification techniques, including artificial neural networks and Wiener series, are compared in terms of measurement accuracy.

Analysis, modelling and classification of geospatial data using machine learning

The research considers the problem of spatial data classification using machine learning algorithms: probabilistic neural networks (PNN) and support vector machines (SVM). As a benchmark model simple k-nearest neighbor algorithm is considered. PNN is a neural network reformulation of well known nonparametric principles of probability density modeling using kernel density estimator and Bayesian optimal or maximum a posteriori decision rules. PNN is well suited to problems where not only predictions but also quantification of accuracy and integration of prior information are necessary. An important property of PNN is that they can be easily used in decision support systems dealing with problems of automatic classification. Support vector machine is an implementation of the principles of statistical learning theory for the classification tasks. Recently they were successfully applied for different environmental topics: classification of soil types and hydro-geological units, optimization of monitoring networks, susceptibility mapping of natural hazards. In the present paper both simulated and real data case studies (low and high dimensional) are considered. The main attention is paid to the detection and learning of spatial patterns by the algorithms applied.

Remaining phosphorus estimated by pedotransfer function

Although the determination of remaining phosphorus (Prem) is simple, accurate values could also be estimated with a pedotransfer function (PTF) aiming at the additional use of soil analysis data and/or Prem replacement by an even simpler determination. The purpose of this paper was to develop a pedotransfer function to estimate Prem values of soils of the State of São Paulo based on properties with easier or routine laboratory determination. A pedotransfer function was developed by artificial neural networks (ANN) from a database of Prem values, pH values measured in 1 mol L-1 NaF solution (pH NaF) and soil chemical and physical properties of samples collected during soil classification activities carried out in the State of São Paulo by the Agronomic Institute of Campinas (IAC). Furthermore, a pedotransfer function was developed by regressing Prem values against the same predictor variables of the ANN-based PTF. Results showed that Prem values can be calculated more accurately with the ANN-based pedotransfer function with the input variables pH NaF values along with the sum of exchangeable bases (SB) and the exchangeable aluminum (Al3+) soil content. In addition, the accuracy of the Prem estimates by ANN-based PTF were more sensitive to increases in the experimental database size. Although the database used in this study was not comprehensive enough for the establishment of a definitive pedotrasnfer function for Prem estimation, results indicated the inclusion of Prem and pH NaF measurements among the soil testing evaluations as promising ind order to provide a greater database for the development of an ANN-based pedotransfer function for accurate Prem estimates from pH NaF, SB, and Al3+ values.

Mult-class differentiation of cannabis seedlings in a forensic context

This article presents an experimental study about the classification ability of several classifiers for multi-classclassification of cannabis seedlings. As the cultivation of drug type cannabis is forbidden in Switzerland lawenforcement authorities regularly ask forensic laboratories to determinate the chemotype of a seized cannabisplant and then to conclude if the plantation is legal or not. This classification is mainly performed when theplant is mature as required by the EU official protocol and then the classification of cannabis seedlings is a timeconsuming and costly procedure. A previous study made by the authors has investigated this problematic [1]and showed that it is possible to differentiate between drug type (illegal) and fibre type (legal) cannabis at anearly stage of growth using gas chromatography interfaced with mass spectrometry (GC-MS) based on therelative proportions of eight major leaf compounds. The aims of the present work are on one hand to continueformer work and to optimize the methodology for the discrimination of drug- and fibre type cannabisdeveloped in the previous study and on the other hand to investigate the possibility to predict illegal cannabisvarieties. Seven classifiers for differentiating between cannabis seedlings are evaluated in this paper, namelyLinear Discriminant Analysis (LDA), Partial Least Squares Discriminant Analysis (PLS-DA), Nearest NeighbourClassification (NNC), Learning Vector Quantization (LVQ), Radial Basis Function Support Vector Machines(RBF SVMs), Random Forest (RF) and Artificial Neural Networks (ANN). The performance of each method wasassessed using the same analytical dataset that consists of 861 samples split into drug- and fibre type cannabiswith drug type cannabis being made up of 12 varieties (i.e. 12 classes). The results show that linear classifiersare not able to manage the distribution of classes in which some overlap areas exist for both classificationproblems. Unlike linear classifiers, NNC and RBF SVMs best differentiate cannabis samples both for 2-class and12-class classifications with average classification results up to 99% and 98%, respectively. Furthermore, RBFSVMs correctly classified into drug type cannabis the independent validation set, which consists of cannabisplants coming from police seizures. In forensic case work this study shows that the discrimination betweencannabis samples at an early stage of growth is possible with fairly high classification performance fordiscriminating between cannabis chemotypes or between drug type cannabis varieties.

Incline, speed, and distance assessment during unconstrained walking.

Body accelerations during human walking were recorded by a portable measuring device. A new method for parameterizing body accelerations and finding the pattern of walking is outlined. Two neural networks were designed to recognize each pattern and estimate the speed and incline of walking. Six subjects performed treadmill walking followed by self-paced walking on an outdoor test circuit involving roads of various inclines. The neural networks were first "trained" by known patterns of treadmill walking. Then the inclines, the speeds, and the distance covered during overground walking (outdoor circuit) were estimated. The results show a good agreement between actual and predicted variables. The standard deviation of estimated incline was less than 2.6% and the maximum of the coefficient of variation of speed estimation is 6%. To the best of our knowledge, these results constitute the first assessment of speed, incline and distance covered during level and slope walking and offer investigators a new tool for assessing levels of outdoor physical activity.

Model-Free Reconstruction of Excitatory Neuronal Connectivity from Calcium Imaging Signals

A systematic assessment of global neural network connectivity through direct electrophysiological assays has remained technically infeasible, even in simpler systems like dissociated neuronal cultures. We introduce an improved algorithmic approach based on Transfer Entropy to reconstruct structural connectivity from network activity monitored through calcium imaging. We focus in this study on the inference of excitatory synaptic links. Based on information theory, our method requires no prior assumptions on the statistics of neuronal firing and neuronal connections. The performance of our algorithm is benchmarked on surrogate time series of calcium fluorescence generated by the simulated dynamics of a network with known ground-truth topology. We find that the functional network topology revealed by Transfer Entropy depends qualitatively on the time-dependent dynamic state of the network (bursting or non-bursting). Thus by conditioning with respect to the global mean activity, we improve the performance of our method. This allows us to focus the analysis to specific dynamical regimes of the network in which the inferred functional connectivity is shaped by monosynaptic excitatory connections, rather than by collective synchrony. Our method can discriminate between actual causal influences between neurons and spurious non-causal correlations due to light scattering artifacts, which inherently affect the quality of fluorescence imaging. Compared to other reconstruction strategies such as cross-correlation or Granger Causality methods, our method based on improved Transfer Entropy is remarkably more accurate. In particular, it provides a good estimation of the excitatory network clustering coefficient, allowing for discrimination between weakly and strongly clustered topologies. Finally, we demonstrate the applicability of our method to analyses of real recordings of in vitro disinhibited cortical cultures where we suggest that excitatory connections are characterized by an elevated level of clustering compared to a random graph (although not extreme) and can be markedly non-local.

Automatic Mapping and Classification of Spatial Environmental Data

The paper deals with the development and application of the generic methodology for automatic processing (mapping and classification) of environmental data. General Regression Neural Network (GRNN) is considered in detail and is proposed as an efficient tool to solve the problem of spatial data mapping (regression). The Probabilistic Neural Network (PNN) is considered as an automatic tool for spatial classifications. The automatic tuning of isotropic and anisotropic GRNN/PNN models using cross-validation procedure is presented. Results are compared with the k-Nearest-Neighbours (k-NN) interpolation algorithm using independent validation data set. Real case studies are based on decision-oriented mapping and classification of radioactively contaminated territories.

Multitask learning of environmental spatial data

The present research deals with an application of artificial neural networks for multitask learning from spatial environmental data. The real case study (sediments contamination of Geneva Lake) consists of 8 pollutants. There are different relationships between these variables, from linear correlations to strong nonlinear dependencies. The main idea is to construct a subsets of pollutants which can be efficiently modeled together within the multitask framework. The proposed two-step approach is based on: 1) the criterion of nonlinear predictability of each variable ?k? by analyzing all possible models composed from the rest of the variables by using a General Regression Neural Network (GRNN) as a model; 2) a multitask learning of the best model using multilayer perceptron and spatial predictions. The results of the study are analyzed using both machine learning and geostatistical tools.

Activation and inhibition of bimanual movements in school-aged children

The development of motor activation and inhibition was compared in 6-to-12 year-olds. Children had to initiate or stop the externally paced movements of one hand, while maintaining that of the other hand. The time needed to perform the switching task (RT) and the spatio-temporal variables show different agerelated evolutions depending on the coordination pattern (inor anti-phase) and the type of transition (activation, selective inhibition, non selective inhibition) required. In the anti-phase mode, activation perturbs the younger subjects' responses while temporal and spatial stabilities transiently decrease around 9 years when activating in the in-phase mode. Aged-related changes differed between inhibition and activation in the antiphase mode, suggesting either the involvement of distinct neural networks or the existence of a single network that is reorganized. In contrast, stopping or adding one hand in the in-phase mode shows similar aged-related improvement. We suggest that selectively stopping or activating one arm during symmetrical coordination rely on the two faces of a common processing in which activation could be the release of inhibition

Remote Synchronization Reveals Network Symmetries and Functional Modules

We study a Kuramoto model in which the oscillators are associated with the nodes of a complex network and the interactions include a phase frustration, thus preventing full synchronization. The system organizes into a regime of remote synchronization where pairs of nodes with the same network symmetry are fully synchronized, despite their distance on the graph. We provide analytical arguments to explain this result, and we show how the frustration parameter affects the distribution of phases. An application to brain networks suggests that anatomical symmetry plays a role in neural synchronization by determining correlated functional modules across distant locations.

Machine learning analysis and modeling of interest rate curves

The present research deals with the review of the analysis and modeling of Swiss franc interest rate curves (IRC) by using unsupervised (SOM, Gaussian Mixtures) and supervised machine (MLP) learning algorithms. IRC are considered as objects embedded into different feature spaces: maturities; maturity-date, parameters of Nelson-Siegel model (NSM). Analysis of NSM parameters and their temporal and clustering structures helps to understand the relevance of model and its potential use for the forecasting. Mapping of IRC in a maturity-date feature space is presented and analyzed for the visualization and forecasting purposes.

Advanced geostatistical and machine-learning models for spatial data analysis of radioactively contaminated regions

Radioactive soil-contamination mapping and risk assessment is a vital issue for decision makers. Traditional approaches for mapping the spatial concentration of radionuclides employ various regression-based models, which usually provide a single-value prediction realization accompanied (in some cases) by estimation error. Such approaches do not provide the capability for rigorous uncertainty quantification or probabilistic mapping. Machine learning is a recent and fast-developing approach based on learning patterns and information from data. Artificial neural networks for prediction mapping have been especially powerful in combination with spatial statistics. A data-driven approach provides the opportunity to integrate additional relevant information about spatial phenomena into a prediction model for more accurate spatial estimates and associated uncertainty. Machine-learning algorithms can also be used for a wider spectrum of problems than before: classification, probability density estimation, and so forth. Stochastic simulations are used to model spatial variability and uncertainty. Unlike regression models, they provide multiple realizations of a particular spatial pattern that allow uncertainty and risk quantification. This paper reviews the most recent methods of spatial data analysis, prediction, and risk mapping, based on machine learning and stochastic simulations in comparison with more traditional regression models. The radioactive fallout from the Chernobyl Nuclear Power Plant accident is used to illustrate the application of the models for prediction and classification problems. This fallout is a unique case study that provides the challenging task of analyzing huge amounts of data ('hard' direct measurements, as well as supplementary information and expert estimates) and solving particular decision-oriented problems.