Net Architecture Research Articles

Sensing, tracking and reasoning with relations

Suppose we have a set of sensor nodes spread over a geographical area. Assume that these nodes are able to perform processing as well as sensing and are additionally capable of communicating with each other by means of a wireless network. Though each node is an independent hardware device, they need to coordinate their sensing, computation and communication to acquire relevant information about their environment so as to accomplish some high-level task. The integration of processing makes such nodes more autonomous and the entire system, which we call a sensor net, becomes a novel type of sensing, processing, and communication engine. The sensor net architecture presented in this article starts from a high-level description of the mission or task to be accomplished and then commands individual nodes to sense and communicate in a manner that accomplishes the desired result with attention to minimizing the computational, communication, and sensing resources required. Much work remains to be done to refine and implement the relational sensing ideas presented here and validate their performance. We believe, however, that the potential pay-off for the relation-based sensing and tracking we have proposed can be large, both in terms of developing rich theories on the design and complexity of sensing algorithms, as well as in terms of the eventual impact of the deployed sensor systems.

System identification using evolutionary Markov chain Monte Carlo

System identification involves determination of the functional structure of a target system that underlies the observed data. In this paper, we present a probabilistic evolutionary method that optimizes system architectures for the identification of unknown target systems. The method is distinguished from existing evolutionary algorithms (EAs) in that the individuals are generated from a probability distribution as in Markov chain Monte Carlo (MCMC). It is also distinguished from conventional MCMC methods in that the search is population-based as in standard evolutionary algorithms. The effectiveness of this hybrid of evolutionary computation and MCMC is tested on a practical problem, i.e., evolving neural net architectures for the identification of nonlinear dynamic systems. Experimental evidence supports that evolutionary MCMC (or eMCMC) exploits the efficiency of simple evolutionary algorithms while maintaining the robustness of MCMC methods and outperforms either approach used alone.

Optimization and empirical modeling of HG-ICP-AES analytical technique through artificial neural networks.

An artificial neural network technique has been applied to the optimization of a hydride generation-inductively coupled plasma-atomic emission spectrometry (HG-ICP-AES) coupling for the determination of Ge at trace levels. The back propagation of errors net architecture was used. Experimental parameters and their relationship have been studied, obtaining a surface response of the system. The results and optimization aspects achieved with the neural network approach have been compared to the "one variable at time" and SIMPLEX methods.

Gamma–hadron discrimination in extensive air showers using a neural network

A neural algorithm was developed to separate electromagnetic and hadronic showers detected with an air shower array. The requirements on the detector performance are very general, so that the results of the calculation can be applied to a wide set of detectors, actually operating or planned for the future. More then 700 000 showers were generated using the Corsika package and were propagated through an ideal pixel-like detector. The peculiarities of each class of showers are presented in detail and it is shown how the neural net architecture is structured around them. The neural net performances were studied for different sets of simulated data. The physics relevance of the gamma–hadron separation is also discussed.

Generalizations of the Hamming Associative Memory

This Letter reviews four models of associative memory which generalize the operation of the Hamming associative memory: the grounded Hamming memory, the cellular Hamming memory, the decoupled Hamming memory, and the two-level decoupled Hamming memory. These memory models offer high performance and allow for a more practical hardware realization than the Hamming net and other fully interconnected neural net architectures.

Weight loss and net abnormalities of Hydropsyche betteni (caddisfly) larvae exposed to aqueous zinc

AbstractCaddisfly larvae (Hydropsyche betteni) were collected near a zinc mining operation and exposed to elevated concentrations of zinc in an attempt to determine the efficacy of using weight change and capture net architecture for assessing the toxicological impact of metal exposure. One group of larvae was collected near the mine site (Adjacent) with another collected upstream and away from the mine (Remote). Weight change and capture net architecture was monitored on 400 individually identifiable larvae. The threshold‐observed‐effect concentration after 13 d was 7.6 mg/L for the Remote group and 30.2 mg/L for the Adjacent group. After 48 d, the highest test concentration of 42 mg/L resulted in a 23 and 19% loss in live‐body weight in the Remote and Adjacent groups, respectively, while control larvae from both groups gained 6% in comparison to initial pre‐exposure weights. Five weeks of exposure at 22 mg Zn/L or greater was required by both larval groups to statistically reduce the frequency of normal nets, indicating that the sensitivity of the net response was less sensitive than weight loss. Relative differences in weight changes and net architecture strongly suggest that the Adjacent larval group was slightly more tolerant than the Remote and that this increased tolerance may be related to chronic pre‐exposure to zinc prior to collection of larvae. This study demonstrates the utility of using weight loss, net architecture, and tissue burdens for assessing the impact of elevated zinc.

A neural network architecture for visual selection.

This article describes a parallel neural net architecture for efficient and robust visual selection in generic gray-level images. Objects are represented through flexible star-type planar arrangements of binary local features which are in turn star-type planar arrangements of oriented edges. Candidate locations are detected over a range of scales and other deformations, using a generalized Hough transform. The flexibility of the arrangements provides the required invariance. Training involves selecting a small number of stable local features from a predefined pool, which are well localized on registered examples of the object. Training therefore requires only small data sets. The parallel architecture is constructed so that the Hough transform associated with any object can be implemented without creating or modifying any connections. The different object representations are learned and stored in a central module. When one of these representations is evoked, it "primes" the appropriate layers in the network so that the corresponding Hough transform is computed. Analogies between the different layers in the network and those in the visual system are discussed. Furthermore, the model can be used to explain certain experiments on visual selection reported in the literature.

Neural networks and statistical inference: seeking robust and efficient learning

Semiparametric statistical inference concepts are briefly reviewed and applied to artificial neural networks. We deal with asymptotic efficiency and robustness aspects of learning, two properties which represent key factors for the quality of the estimates that statisticians obtain. In particular, the accuracy of the estimates in the presence of outlying observations is an important goal since it is a well-known fact that between efficiency and robustness one seeks the best compromise. With that scope in mind, we analyze possible ways of building up net architectures without relying on strong assumptions about the functional components in the model.

Performance Evaluation of Neural Networks in Concrete Condition Assessment

A neural network modeling approach is used to identify concrete specimens that contain internal cracks. Different types of neural nets are used and their performance is evaluated. Correct classification of the signals received from a cracked specimen could be achieved with an accuracy of 75 percent for the test set and 95 percent for the training set. These recognition rates lead to the correct classification of all the individual test specimens. Although some neural net architectures may show high performance with a particular training data set, their results might be inconsistent. In situations in which the number of data sets is small, consistent performance of a neural network may be achieved by shuffling the training and testing data sets.

Potential applications of artificial neural networks to thermodynamics: vapor–liquid equilibrium predictions

The associative property of artificial neural networks (ANNs) and their inherent ability to “learn” and “recognize” highly non-linear and complex relationships finds them ideally suited to a wide range of applications in chemical engineering. Dynamic Modeling and Control of Chemical Process Systems and Fault Diagnosis are the two significant applications of ANNs that have been explored so far with success. This paper deals with the potential applications of ANNs in thermodynamics — particularly, the prediction/estimation of vapor–liquid equilibrium (VLE) data. The prediction of VLE data by conventional thermodynamic methods is tedious and requires determination of “constants” which is arbitrary in many ways. Also, the use of conventional thermodynamics for predicting VLE data for highly polar substances introduces a large number of inaccuracies. The possibility of applying ANNs for VLE data prediction/estimation has been explored using the back propagation algorithm. The methane–ethane and ammonia–water systems have been studied and the VLE predictions have been found to be accurate to within ±1%. Preliminary results confirm exciting possibilities of ANNs for applications to thermodynamics of mixtures. Advantages and limitations of this application are also discussed. An heuristic approach to reduce the trial and error process for selecting the “optimum” net architecture is discussed.

Role of the net architecture in piriform cortex activity: analysis by a mathematical model.

We present a mathematical analysis of the piriform cortex activity in rats. Experimental data were obtained by means of optical recording of fluorescent signals driven by neuronal activity. From these data, we determined the numerical value of the relaxation time for the pyramidal cell activity in layers II and III and the time latency map for bulb activation. Our model for the piriform cortex is based on pairs of excitatory and inhibitory neurons which correspond to pyramidal cells of layers II and III and to their inhibitory associated interneurons respectively; pyramidal cells are also interconnected through short and long range association fiber systems. Under such conditions, the model outputs resemble closely the experimental observations: (1) a double-bumped response to a strong and short stimulation; (2) oscillatory behavior under weak sustained stimulation conditions; (3) propagation of traveling activity waves; and (4) pacemaker activity when clusters of neurons are preferentially coupled.

Pattern recognition models for spectral reflectance evaluation of apple blemishes

Abstract Surface blemishes of various apple varieties were analyzed by their reflectance characteristics between 460 and 1130 nm. Normalized reflectance data were collected at 10 nm increments with liquid crystal tunable filters. Data were utilized as input values for various pattern recognition models specifically multi-layer back propagation, unimodal Gaussian, K-nearest neighbor and nearest cluster algorithms. Partitioning data into 50:50 training and test sets, correct classification in separating unflawed versus blemished areas ranged from 62 to 96% (Year I) and from 73 to 85% (Year II). The algorithm which yielded the highest correct classification was the multi-layer back propagation but minor variation was found for number of hidden nodes or neural net architecture.

Markov chain Monte Carlo methods for probabilistic network model determination

The predictive performance of a neural network model depends crucially on the net architecture. Although complex models can fit very well the data at hand, their out-of-sample performance is typically very poor. In order to improve the predictive performance of a neural network model, we propose to interpret a neural network as a graphical model with latent variables. The proposed interpretation allows to extend a computational methodology recently proposed in Giudici and Green (1999) to compare alternative network models in terms of their posterior probabilities, and, therefore, to choose the most plausible net architecture. We apply our proposal to a data-set concerned with the analysis of the correlation structure among shares of viewers in the Italian television market.

Grammatical analysis of Hindi sentences using connectionist approach

Abstract: This paper presents a simple connectionist approach to parsing of a subset of sentences in the Hindi language, using Rule based Connectionist Networks (RBCN) as suggested by Fu in 1993. The basic grammar rules representing Kernel Hindi sentences have been used to determine the initial topology of the RBCN. The RBCN is based on a multilayer perceptron, trained using the backpropagation algorithm. The terminal symbols defined in the language structure are mapped onto the input nodes, the non‐terminals onto hidden nodes and the start symbol onto the single output node of the network structure. The training instances are sentences of arbitrary, but fixed maximum length and fixed word order. A neural network based recognizer is used to perform grammaticality determination and parse tree generation of a given sentence. The network is exposed to both positive and negative training instances, derived from a simple context‐free‐grammar (CFG), during the training phase. The trained network recognizes seen sentences (sentences present in the training set) with 98–100% accuracy. Since a neural net based recognizer is trainable in nature, it can be trained to recognize any other CFG, simply by changing the training set. This results in reducing programming effort involved in parser development, as compared to that of the conventional AI approach. The parsing time is also reduced to a great extent as compared to that of a conventional parser, as a result of the inherent parallelism exhibited by neural net architecture.

Local cluster neural net: Architecture, training and applications

This paper describes the structure, training and computational abilities of the local cluster (LC) artificial neural net architecture. LC nets are a special class of multilayer perceptrons that use sigmoid functions to generate localised functions. LC nets train as fast as radial basis functions nets and are more general. They are well suited for both, multi-dimensional function approximation and discrete classification. The LC net is the result of our search for a widely applicable neural net architecture suitable for low-cost hardware realisation. The LC net seem particularly well suited for analog VLSI realisation of small-size, low-power, fully parallel neural net chip for real time control applications.

Automatic Classification of Steels by Processing Energy-Dispersive X-ray Spectra with Artificial Neural Networks

The automatic classification of steels has been studied. The chemical compositions of 19 certificate steels were correlated with its energy dispersion X-ray fluorescence spectra. Twelve relevant elements of these samples were selected for data processing through artificial neural networks (ANNs). A Kohonen type ANN of 8 × 8 × 1 1 dimension was used. This net architecture allows on-line classification with 100% efficiency, that is, without errors.

Extranet Security: What's Right for the Business?

Before traveling too far down the path of presenting secure extra- net architectures, it is helpful to get a solid foundation of what an extranet actually is. In and of itself, this discussion could provide sufficient material for a dissertation; however, to keep it brief, any omission of an opinion is clearly intentional.

An adaptive algorithm for speech analysis and synthesis.

In this paper, an artificial neural net architecture for adaptive speech analysis and synthesis is proposed. Based on discussing the signal analysis procedure with mathematical method, the simulation experiment is performed firstly on a typical signal (a square wave). Compared with Fast Fourier Transform, some important characteristics of the Neural Net Signal Analyzer-Synthesizer (NNSAS) are drawn from the experiment results. Then, NNSAS is used for speech signal processing, which results in a new approach to speech analysis and synthesis. In terms of the auditory masking effect, only some major components of connection strength vector are used to produce synthetic speech. Under the high compressive rate conditions, NNSAS can still produce synthetic speech with high articulation.

Feedforward neural nets as discretization schemes for ODEs and DAEs

Because of their neurophysical origin neural nets can be studied for classification tasks, approximation properties or iterative algorithms. They can be interpreted as a distributed or massively parallel computer, where each unit accumulates a scalar product and computers an one-dimensional nonlinear activation function. A supervised learning strategy defines a nonlinear least-squares problem, which is solved by gradient techniques like backpropagation. Interpreting the weights in a net as state variables, feedforward neural nets can be designed as numerical discretization schemes for ODEs or DAEs. We present the net architecture for the implicit Euler scheme and solve some test examples numerically. The net approach is especially of interest for the overdetermined index-3 approach of DAEs from multibody system dynamics. In general, these nets define a parallel shooting-type algorithm. Its merits are in real-time applications, since a hardware realization of the net is possible.

Simulation of hydraulic power plant transients using neural networks

In this work the efficiency of the artificial neural network technology for evaluating the turbine speed and drain synchronous valve during the turn-off transient phase of an actual hydraulic power plant has been evaluated. These results have been achieved avoiding a detailed time consuming analysis of the power plant. Several net architectures have been set up and verified with regard to their response to this strongly non-linear problem.