Investigation of multilingual deep neural networks for spoken term detection

The development of high-performance speech processing systems for low-resource languages is a challenging area. One approach to address the lack of resources is to make use of data from multiple languages. A popular direction in recent years is to use bottleneck features, or hybrid systems, trained on multilingual data for speech-to-text (STT) systems. This paper presents an investigation into the application of these multilingual approaches to spoken term detection. Experiments were run using the IARPA Babel limited language pack corpora (∼10 hours/language) with 4 languages for initial multilingual system development and an additional held-out target language. STT gains achieved through using multilingual bottleneck features in a Tandem configuration are shown to also apply to keyword search (KWS). Further improvements in both STT and KWS were observed by incorporating language questions into the Tandem GMM-HMM decision trees for the training set languages. Adapted hybrid systems performed slightly worse on average than the adapted Tandem systems. A language independent acoustic model test on the target language showed that retraining or adapting of the acoustic models to the target language is currently minimally needed to achieve reasonable performance. © 2013 IEEE.

Modelling algal blooms in Lake Dianchi, China using neural networks

Lake Dianchi is one of the most extensively impacted freshwater lakes by algal blooms. To investigate the response of dominant algal genera, neural networks were applied to model the relationship between water quality parameters and the biomass of four dominant genera (Microcystic spp., Anabaena sp., Quadricauda (Turp.) Breb, Pediastrum Mey) in Dianchi. Results showed that the timing and magnitude of algal blooms of Microcystic spp., nabaena sp., Quadricauda (Turp.) Breb, and Pediastrum Mey in Dianchi could be successfully predicted. The evaluation of environmental factors showed that pH had more significant impact on concentrations of all the four dominant algal genera than the nutrient factors, such as total phosphorus and total nitrogen.

Single-electron tunneling depressing synapse for cellular neural networks

In this paper, a cellular neural network with depressing synapses for contrast-invariant pattern classification and synchrony detection is presented, starting from the impulse model of the single-electron tunneling junction. The results of the impulse model and the network are simulated using simulation program with integrated circuit emphasis (SPICE). It is demonstrated that depressing synapses should be an important candidate of robust systems since they exhibit a rapid depression of excitatory postsynaptic potentials for successive presynaptic spikes.

The application of DBF neural networks for object recognition

On the basis of DBF nets proposed by Wang Shoujue, the model and properties of DBF neural network were discussed in this paper. When applied in pattern recognition, the algorithm and implement on hardware were presented respectively. We did experiments on recognition of omnidirectionally oriented rigid objects on the same level, using direction basis function neural networks, which acts by the method of covering the high dimensional geometrical distribution of the sample set in the feature space. Many animal and vehicle models (even with rather similar shapes) were recognized omnidirectionally thousands of times. For total 8800 tests, the correct recognition rate is 98.75%, the error rate and the rejection rate are 0.5% and 1.25% respectively. (C) 2003 Elsevier Inc. All rights reserved.

A novel face recognition method based on hypersausage neural networks

In this paper, we firstly give the nature of 'hypersausages', study its structure and training of the network, then discuss the nature of it by way of experimenting with ORL face database, and finally, verify its unsurpassable advantages compared with other means.

Cancer classification based on direction-basis-function neural networks

Automatic molecular classification of cancer based on DNA microarray has many advantages over conventional classification based on morphological appearance of the tumor. Using artificial neural networks is a general approach for automatic classification. In this paper, Direction-Basis-Function neuron and Priority-Ordered algorithm are applied to neural networks. And the leukemia gene expression dataset is used as an example to testify the classifier. The result of our method is compared to that of SVM. It shows that our method makes a better performance than SVM.

Kinetic Model Study on Enzymatic Hydrolysis of Cellulose Using Artificial Neural Networks

Enzymatic hydrolysis of cellulose was highly complex because of the unclear enzymatic mechanism and many factors that affect the heterogeneous system. Therefore, it is difficult to build a theoretical model to study cellulose hydrolysis by cellulase. Artificial neural network (ANN) was used to simulate and predict this enzymatic reaction and compared with the response surface model (RSM). The independent variables were cellulase amount X-1, substrate concentration X-2, and reaction time X-3, and the response variables were reducing sugar concentration Y-1 and transformation rate of the raw material Y-2. The experimental results showed that ANN was much more suitable for studying the kinetics of the enzymatic hydrolysis than RSM. During the simulation process, relative errors produced by the ANN model were apparently smaller than that by RSM except one and the central experimental points. During the prediction process, values produced by the ANN model were much closer to the experimental values than that produced by RSM. These showed that ANN is a persuasive tool that can be used for studying the kinetics of cellulose hydrolysis catalyzed by cellulase.

Artificial neural networks expecting more advanced microelectronics technology

　

Global stability of Hopfield neural networks

This paper gives a condition for the global stability of a continuous-time hopfield neural network when its activation function maybe not monotonically increasing.

Classification of valence changes of trivalent rare earth ions in alkaline earth borates using artificial neural networks

The investigations of classification on the valence changes from RE3+ to RE2+ (RE = Eu, Sm, Yb, Tm) in host compounds of alkaline earth berate were performed using artificial neural networks (ANNs). For comparison, the common methods of pattern recognition, such as SIMCA, KNN, Fisher discriminant analysis and stepwise discriminant analysis were adopted. A learning set consisting of 24 host compounds and a test set consisting of 12 host compounds were characterized by eight crystal structure parameters. These parameters were reduced from 8 to 4 by leaps and bounds algorithm. The recognition rates from 87.5 to 95.8% and prediction capabilities from 75.0 to 91.7% were obtained. The results provided by ANN method were better than that achieved by the other four methods. (C) 1999 Elsevier Science B.V. All rights reserved.

A Theory of Networks for Appxoimation and Learning

Learning an input-output mapping from a set of examples, of the type that many neural networks have been constructed to perform, can be regarded as synthesizing an approximation of a multi-dimensional function, that is solving the problem of hypersurface reconstruction. From this point of view, this form of learning is closely related to classical approximation techniques, such as generalized splines and regularization theory. This paper considers the problems of an exact representation and, in more detail, of the approximation of linear and nolinear mappings in terms of simpler functions of fewer variables. Kolmogorov's theorem concerning the representation of functions of several variables in terms of functions of one variable turns out to be almost irrelevant in the context of networks for learning. We develop a theoretical framework for approximation based on regularization techniques that leads to a class of three-layer networks that we call Generalized Radial Basis Functions (GRBF), since they are mathematically related to the well-known Radial Basis Functions, mainly used for strict interpolation tasks. GRBF networks are not only equivalent to generalized splines, but are also closely related to pattern recognition methods such as Parzen windows and potential functions and to several neural network algorithms, such as Kanerva's associative memory, backpropagation and Kohonen's topology preserving map. They also have an interesting interpretation in terms of prototypes that are synthesized and optimally combined during the learning stage. The paper introduces several extensions and applications of the technique and discusses intriguing analogies with neurobiological data.

Endocrine Inspired Modulation of Artificial Neural Networks for Mobile Robotics

Sauze, C and Neal, M. 'Endocrine Inspired Modulation of Artificial Neural Networks for Mobile Robotics', Dynamics of Learning Behavior and Neuromodulation Workshop, European Conference on Artifical Life 2007, Lisbon, Portugal, September 10th-14th 2007.

Rough set feature selection and rule induction for prediction of malignancy degree in brain glioma

X. Wang, J. Yang, R. Jensen and X. Liu, 'Rough Set Feature Selection and Rule Induction for Prediction of Malignancy Degree in Brain Glioma,' Computer Methods and Programs in Biomedicine, vol. 83, no. 2, pp. 147-156, 2006.