Training Rule Research Articles

Synthesis of Polysyllabic Sequences of Thai Tones Using a Generative Model of Fundamental Frequency Contours

In this paper, the distinctive tones of Thai in running speech are studied. We present rules to synthesize F0 contours of Thai tones in running speech by using the generative model of F0 contours. Along with our method, the pitch contours of Thai polysyllabic words, both disyllabic and trisyllabic words, were analyzed. The coarticulation effect of Thai tones in running speech were found. Based on the analysis of the polysyllabic words using this model, rules are derived and applied to synthesize Thai polysyllabic tone sequences. We performed listening tests to evaluate intelligibility of the rules for Thai tones generation. The average intelligibility scores became 98.8%, and 96.6% for disyllabic and trisyllabic words, respectively. From these result, the rule of the tones' generation was shown to be effective. Furthermore, we constructed the connecting rules to synthesize suprasegmental F0 contours using the trisyllable training rules' parameters. The parameters of the first, the third, and the second syllables were selected and assigned to the initial, the ending, and the remaining syllables in a sentence, respectively. Even such a simple rule, the synthesized phrases/senetences were completely identified in listening tests. The MOSs (Mean Opinion Score) was 3.50 while the original and analysis/synthesis samples were 4.82 and 3.59, respectively.

The evaluation of parameters affecting accidents in companies using industrial radioactive sources in Iran.

More than 1000 radiation workers are engaged in about 70 private industrial radiography companies (IRCs) using 312 mobile gamma radiography devices (GRDs) in Iran. These companies mainly carry out radiography in open areas using 192Ir radioactive sources, ranging from 0.07 to 3 TBq (2-80 Ci). The radiation accidents that occurred in these companies led to acute radiation syndrome in 13 workers and overexposure of 36 workers since 1998. The facilities and operating procedures of the Iranian private IRCs have been investigated during the past 5 y. According to this study, in spite of the fact that most of the accidents were found to be due to the failure of critical components of GRDs, the main parameters affecting overexposures and deterministic effects were found to be incorrect or there was insufficient implementation of the monitoring programme, training and local rules in the IRCs.

Classification capacity of a modular neural network implementing neurally inspired architecture and training rules.

A three-layer neural network (NN) with novel adaptive architecture has been developed. The hidden layer of the network consists of slabs of single neuron models, where neurons within a slab--but not between slabs--have the same type of activation function. The network activation functions in all three layers have adaptable parameters. The network was trained using a biologically inspired, guided-annealing learning rule on a variety of medical data. Good training/testing classification performance was obtained on all data sets tested. The performance achieved was comparable to that of SVM classifiers. It was shown that the adaptive network architecture, inspired from the modular organization often encountered in the mammalian cerebral cortex, can benefit classification performance.

Relationship of occupational injuries with social and economic factors

To explore the relationship of occupational injuries with social and economic factors in chemical industry during 2000.01 - 2001.12. 1:2 paired case-control study, univariable logistic regression analysis, principal component analysis, and multiple logistic regression analysis were used in this study. Univariable analysis showed that occupational injuries had significant relationship with age, sex, education, employment pattern, technology, workplace, work changing, wage, family income, enterprise scale, enterprise proprietorship, projective device, operation rules, and training rules of work safety. The extracted four principal components (PC(1), PC(2), PC(3) and PC(4), ranked by contribution) gave good expressions to the initial 11 variables. The cumulative proportion of the four principal components reached 77.36%. PC(1) was the indicative factor of occupational injuries, which represented 46.69% information of initial variables. PC(2) was the kinetic factor of occupational injuries. PC(3) was the stable factor of occupational injuries. PC(4) was the sex factor of occupational injuries. The results of multiple conditional logistic regression analysis showed that occupational injuries had statistically significant relationship with PC(1) and PC(2). Among the initial variables, sex, employment pattern, income, scale of enterprise, and property of enterprise were more prominent. Occupational injuries are related with multiple social and economic factors, which often interact on each other. The prevention and control of occupational injuries should require a comprehensive approach, including training and education of work safety, improving workers' consciousness of self-protection, and enhancing proprietors' consciousness of work safety.

Toward Intelligent Variable Message Signs in Freeway Work Zones: Neural Network Model

An increasingly popular method of managing freeway traffic is to use variable message signs (VMS). A neural network model is presented for real-time control of a VMS system in freeway work zones. The neural network is trained to detect the start of a queue in a work zone and provide a message in the freeway upstream. The travelers are informed about the congestion in a work zone when a queue starts to form. The intelligent VMS system can be trained with data for different periods within a day, such as morning and evening rush hours, nonrush hours during the day, and night, for a more detailed traffic flow prediction over the period of one day. Two different neural network training rules are used: the simple backpropagation (BP) and the Levenberg–Marquardt BP algorithms. The network is trained using data adapted from the measured data. Based on different numerical experiments it is observed that the convergence speed of the Levenberg–Marquardt BP algorithm is at least one order of magnitude faster than the simple BP algorithm for the work zone traffic queue detection problem.

Generalized rules for combination and joint training of classifiers

Classifier combination has repeatedly been shown to provide significant improvements in performance for a wide range of classification tasks. In this paper, we focus on the problem of combining probability distributions generated by different classifiers. Specifically, we present a set of new combination rules that generalize the most commonly used combination functions, such as the mean, product, min, and max operations. These new rules have continuous and differentiable forms, and can thus not only be used for combination of independently trained classifiers, but also as objective functions in a joint classifier training scheme. We evaluate both of these schemes by applying them to the combination of phone classifiers in a speech recognition system. We find a significant performance improvement over previously used combination schemes when jointly training and combining multiple systems using a generalization of the product rule.

A hazai felnőttképzés jellemzői

The study surveys the most important aspects of the national adult training from educational economy point of view. The authors examine the national limits of legal rule of adult training. They present the different sub-systems of adult training i.e. the educational system, the non-educational system and the working place training characteristics of participation making international comparison too. We look over the target groups of adult-training and the participants motivation of the teaching arena. We also have a look at the domestic costs of the different froms of adult-training, and the division of this expenditure among the participants.

Dansk idræt og den olympiske bevægelse 1894-1932

Dansk idræts forhold til den internationale olympiske bevægelse i omtalte periode.Danish sport and the Olympic movement 1894-1932 The International Olympic Committee (IOC) was formed in June 1894 by Pierre de Coubertin, partly inspired by the international movement for peace. Members were men particularly of the upper classes hand-picked by Coubertin. The Danish Sport Association (DIF) was created by men in 1896 from the Copenhagen bourgeoisie, many of them inspired by England. After about 1900 the working class became a majority in DIF, but the bourgeoisie continued to retain power for many years. Scandinavian sporting collaboration at the organizational level took shape after a couple of failed attempts and became in 1919 “De nordiske Rigs-Idrætsforbunds Fælleskomité” (The Joint Committee of Scandinavian National Sports Federations). The work of the “joint committee” is an example of the way in which a group of minor states, by dint of formalized, organized work, attempted to acquire influence on the IOC during this period. Around the period of the 1st World War, rules for sports training had been left by the IOC to the international special federations, which had now been established. These special federations had also made their own rules for amateurs, the contents of which could, therefore, vary somewhat between one sporting branch and another. However, while the IOC had ceased getting involved in professional sports rules, this was not the case as concerned amateur rules. Here the IOC attempted right up until 1952 to retain its role as watchdog. This has to be seen in the context of the IOC’s view that sport is a educational means to create a better world in which peace and international understanding prevail. This had not originally been Coubertin’s intention with his educational programme, but here the Zeitgeist was against him and he had to give way to ideals of the English gentleman and American middle class attitudes. In Denmark the ideas of Olympism did not catch on. In the period 1897-1918 DIF introduced completely unique amateur rules, which used the term “non-amateur” instead of “professional”. This made it possible to hand out punishments for infringements of “ethical rules” instead of only being able to punish for “economic” infringements as was the custom.

Second-order asymmetric BAM design with a maximal basin of attraction

Bidirectional associative memory (BAM) generalizes the associative memory (AM) to be capable of performing two-way recalling of pattern pairs. Asymmetric bidirectional associative memory (ABAM) is a variant of BAM relaxed with connection weight symmetry restriction and enjoys a much better performance than a conventional BAM structure. Higher-order associative memories (HOAMs) are reputed for their higher memory capacity than the first-order counterparts. The paper concerns the design of a second-order asymmetric bidirectional associative memory (SOABAM) with a maximal basin of attraction, whose extension to a HOABAM is possible and straightforward. First, a necessary and sufficient condition is derived for the connection weight matrix of SOABAM that can guarantee the recall of all prototype pattern pairs. A local training rule which is adaptive in the learning step size is formulated. Then derived is a theorem, designing a SOABAM further enlarging the quantities required to meet the complete recall theorem will enhance the capability of evolving a noisy pattern to converge to its association pattern vector without error. Based on this theorem, our algorithm is also modified to ensure each training pattern is stored with a basin of attraction as large as possible.

Robust H//sub∞/ controller design with recurrent neural network for linear synchronous motor drive

In this paper, a robust controller design with H/sub /spl infin// performance using a recurrent neural network (RNN) is proposed for the position tracking control of a permanent-magnet linear synchronous motor. The proposed robust H/sub /spl infin// controller, which comprises a RNN and a compensating control, is developed to reduce the influence of parameter variations and external disturbance on system performance. The RNN is adopted to estimate the dynamics of the lumped plant uncertainty, and the compensating controller is used to eliminate the effect of the higher order terms in Taylor series expansion of the minimum approximation error. The tracking performance is ensured in face of parameter variations, external disturbance and RNN estimation error once a prespecified H/sub /spl infin// performance requirement is achieved. The synthesis of the RNN training rules and compensating control are based on the solution of a nonlinear H/sub /spl infin// control problem corresponding to the desired H/sub /spl infin// performance requirement, which is solved via a choice of quadratic storage function. The proposed control method is able to track both the periodic step and sinusoidal commands with improved performance in face of large parameter perturbations and external disturbance.

Improved Training Rules for Multilayered Feedforward Neural Networks

We propose an improved supervisory training rule for multilayered feedforward neural networks (FNNs). The proposed method analytically estimates the optimal solutions for the output weights of FNNs. Also, using the optimal solutions, it reduces the searching space as much as the output weights in the iterative high-dimensional nonlinear optimization problem for the supervisory training. As a result, we can secure a much faster convergence rate and better robustness compared to the previous full-dimensional training rules.

A new approach to perceptron training

The training of perceptrons is discussed in the framework of nonsmooth optimization. An investigation of Rosenblatt's perceptron training rule shows that convergence or the failure to converge in certain situations can be easily understood in this framework. An algorithm based on results from nonsmooth optimization is proposed and its relation to the "constrained steepest descent" method is investigated. Numerical experiments verify that the "constrained steepest descent" algorithm may be further improved by the integration of methods from nonsmooth optimization.

Neural Networks and Regional Science Modeling: A Survey of Techniques for Complex Spatial Analysis

This study examines the value of utilizing neural net modeling for issues relating to optimization across a network of cities in space. Neural nets are made up of many nonlinear computational elements that operate in parallel and are arranged in a manner similar to biological neural nets. Defining a neural net model involves specifying a net topology, arrangement of nodes, training or learning rules, adjustment of weights associated with connections, node characteristics, and rules of transformation from input to output. All of these are the major issues in such regional problems as labor force migration and firm location.

Generalized neural trees for pattern classification

In this paper, a new neural tree (NT) model, the generalized NT (GNT), is presented. The main novelty of the GNT consists in the definition of a new training rule that performs an overall optimization of the tree. Each time the tree is increased by a new level, the whole tree is reevaluated. The training rule uses a weight correction strategy that takes into account the entire tree structure, and it applies a normalization procedure to the activation values of each node such that these values can be interpreted as a probability. The weight connection updating is calculated by minimizing a cost function, which represents a measure of the overall probability of correct classification. Significant results on both synthetic and real data have been obtained by comparing the classification performances among multilayer perceptrons (MLPs), NTs, and GNTs. In particular, the GNT model displays good classification performances for training sets having complex distributions. Moreover, its particular structure provides an easily probabilistic interpretation of the pattern classification task and allows growing small neural trees with good generalization properties.

Intelligent locomotion control on sloping surfaces

A learning scheme is developed to generate robotic locomotion on different sloping surfaces. This scheme uses three neural networks: a neural network controller, a neural network emulator, and a slope information neural network. The neural network controller is pre-trained by a reference trajectory on a horizontal surface. The emulator is pre-trained to identify the robotic dynamics. The slope information neural network provides compensated control signals to the robot on different slope angles by using the control signals on the horizontal surface from the pre-trained controller. The training rule is a backpropogation algorithm with time delay. The proposed technique can generate gaits on different sloping surfaces by following a reference trajectory with desired step length, crossing height and walking speed.

On the generative probability density model in the self-organizing map

The self-organizing map (SOM), is a widely used tool in exploratory data analysis. A major drawback of the SOM has been the lack of a theoretically justified criterion for model selection. Model complexity has a decisive effect on the reliability of visual data analysis, which is a main application of the SOM. In particular, independence of variables cannot be observed unless generalization of the model is good. We describe the maximum likelihood probability density model which follows from the SOM training rule, and show how the density model can be applied to choose the correct model complexity, based on the method of maximum likelihood.

A field-of-view generation algorithm using neural network

In various works using vision system, if a field of view (FOV) of CCD camera cannot cover overall working area on a table or a floor, the overall working area must be divided into several areas whose sizes equal that of FOV. In this study, an effective FOV generation algorithm in which FOV is automatically made is proposed for vision system. Instead of existing sequential FOV generation method, we propose a new FOV generation method by using newly modified self-organizing map (SOM) which has multiple structure based on SOM of neural network and uses new training rule by introducing the movement, creation and deletion terms. In consequence of real experiment, we show that the proposed algorithm is able to decrease the number of created FOVs about 10% more than the existing sequential algorithm. This result is acquired by which the proposed algorithm minimizes overlaps between FOVs, assigns optimal positions of FOVs and eliminates the unnecessary FOVs. Consequently, as the number of FOVs decreases, we have been able to diminish the total working time using vision system.

Rule Extraction from Fuzzy Neural Networks FuNN: A Method and a Real-World Application

The paper presents a method for rule extraction from fuzzy neural networks FuNN. The method is applied to a real problem - adaptive control of a wastewater treatment reactor The method extracts from a trained FuNN fuzzy rules that have coefficients of importance and certainty factors attached to their condition and conclusion parts respectively. The rules can have different levels of abstraction depending on the chosen values of several parameters. General rules can be used for explanation and understanding purposes, while precise and specific rules can be used for inference over new data.

Pattern recognition in damaged neural networks

We have studied the effect of various kinds of damaging that may occur in a neural network whose synaptic bonds have been trained (before damaging) so as to preserve a definite number of patterns. We have used the Hopfield model of the neural network, and applied the Hebbian rule of training (learning). We have studied networks with 600 elements (neurons) and investigated several types of damaging, by performing very extensive numerical investigation. Thus, we have demonstrated that there is no difference between symmetric and asymmetric damaging of bonds. Besides, it turns out that the worst damaging of synaptic bonds is the one that starts with ruining the strongest bonds, whereas in the opposite case, that is, in the case of damaging that starts with ruining the weakest bonds, the learnt patterns remain preserved even for a large percentage of extinguished bonds.

Neural-morphological approach for pattern classification

In cluster analysis, the mode boundaries are a very important part of the hierarchy of structures that link raw data with their interpretation. The existing mode boundary detection approaches for clustering are conditioned by the adjustment of some parameters, which become critical for large dimensionality data sets. Mode boundary detection can be greatly facilitate by mapping, as a first step of process understanding, a reduction of data dimensionality. Under this assumption, an approach is discussed, based on both neural network and mathematical morphology. It requires neither a starting classification, nor an a priori number of clusters or their distribution. Data projection mapping is done using a multilayer neural network with a fast training rule based on a conjugate gradient. Mode boundaries of the underlying probability density function, estimated from the patterns in the projection space, are then easily obtained by making concepts of morphological watershed transformations suitable for their detection. The observations in the raw data space corresponding to those falling in the so-detected mode boundaries are taken as prototypes for classification. The clustering scheme, illustrated using an artificial simulation, has been applied to determine the clusters inside a set of biometrical six-dimensional data of the Guadeloupe honeybee's races.