Abstract: A neural network system is provided that models the system in a system model (12) with the output thereof providing a predicted output. This predicted output is modified or controlled by an output control (14). Input data is processed in a data preprocess step (10) to reconcile the data for input to the system model (12). Additionally, the error resulted from the reconciliation is input to an uncertainty model to predict the uncertainty in the predicted output. This is input to a decision processor (20) which is utilized to control the output control (14). The output control (14) is controlled to either vary the predicted output or to inhibit the predicted output whenever the output of the uncertainty model (18) exceeds a predetermined decision threshold, input by a decision threshold block (22).

Abstract: A variational inference engine for probabilistic graphical models is disclosed. In one embodiment, a method includes inputting a specification for a model that has observable variables and unobservable variables. The specification includes a functional form for the conditional distributions of the model, and a structure for a graph of model that has nodes for each of the variables. The method determines a distribution for the unobservable variables that approximates the exact posterior distribution, based on the graph's structure and the functional form for the model's conditional distributions. The engine thus allows a user to design, implement and solve models without mathematical analysis or computer coding.

Abstract: An automated method for modeling spectral data is provided, wherein the spectral data generated by one of diffuse reflectance, clear transmission, or diffuse transmission. The method includes accessing a set of spectral data, the set of spectral data including, corresponding spectral data for each of a plurality of samples, the spectral data for each of the plurality of samples having associated therewith at least one constituent value, the at least one constituent value being a reference value for a target substance in the sample which is measured by a independent measurement technique. A plurality of data transforms are applied to the set of spectral data to generate, for each sample, a set of transformed and untransformed spectral data.

Abstract: An information processing system having signal processors that are interconnected by processing junctions that simulate and extend biological neural networks. Each processing junction receives signals from one signal processor and generates a new signal to another signal processor. The response of each processing junction is determined by internal junction processes and is continuously changed with temporal variation in the received signal. Different processing junctions connected to receive a common signal from a signal processor respond differently to produce different signals to downstream signal processors. This transforms a temporal pattern of a signal train of spikes into a spatio-temporal pattern of junction events and provides an exponential computational power to signal processors. Each signal processing junction can receive a feedback signal from a downstream signal processor so that an internal junction process can be adjusted to learn certain characteristics embedded in received signals.

Abstract: An apparatus and method for sensor signal prediction and for improving sensor signal response time, is disclosed. An adaptive filter or an artificial neural network is utilized to provide predictive sensor signal output and is further used to reduce sensor response time delay.

Abstract: A fatigue monitoring system and method is disclosed in which a stream of data relating to the stresses experienced at a plurality of locations over the structure during operation is applied to a neural network trained to remove data stream values deemed to be in error. The data from the neural network is then processed to determine the fatigue life.

Abstract: There is proposed a methodology for specifying the behavior of reactive systems, which is based on “Playing in” the system's possible scenarios. The use of this methodology is shown by building a practical “Play In Engine” tool based on the methodology. Users will be able to connect their Mock-Up GUI to the tool by “playing” their GUI and specifying the required system reactions in a simple and intuitive manner. As this is being done, the system will automatically generate behavior specifications in the language of Live Sequence Charts (LSCs), or any other suitable requirement language, such as various temporal logics or timing diagrams. On the basis of the system behavior specification, the user can play out scenarios through a play out engine.

Abstract: In an information filtering method, attributes included in information items are extracted and stored, and ratings relative to the information items carried out by users are stored. The users include a subject user and other users. A relationship between the ratings relative to the information items rated by the subject user and the attributes thereof and a relationship between the ratings relative to the information items rated by the other users and the attributes thereof are utilized for estimating relevances to the subject user of the information items not rated by the subject user. The estimated relevances are used to carry out recommendation or filtering-in of the information item which matches with the subject user.

Abstract: In a recording/playback system, increased information is achieved by 4 level biased magnetic recording where the maximum amplitude 4 level recording signal drives the medium's magnetization into a nonlinear region of its transfer function. The bias does not eliminate distortion at the maximum signal input level, however the system's signal to noise ratio is improved due to an increase in the amplitude of the playback signal resulting from the increased recording level. The nonlinear mapping capability of a neural network provides equalization of playback signals distorted due to the record/playback nonlinearity. The 4 level recorded signals provide a factor of 2 in information storage compared to binary recording, and quadrature amplitude modulation (QAM) combined with the 4 level recording technique provides an additional factor of 2, for a factor of 4 in the information content stored.

Abstract: Described herein is a method and system for training nonlinear adaptive filters (or neural networks) which have embedded memory. Such memory can arise in a multi-layer finite impulse response (FIR) architecture, or an infinite impulse response (IIR) architecture. We focus on filter architectures with separate linear dynamic components and static nonlinear components. Such filters can be structured so as to restrict their degrees of computational freedom based on a priori knowledge about the dynamic operation to be emulated. The method is detailed for an FIR architecture which consists of linear FIR filters together with nonlinear generalized single layer subnets. For the IIR case, we extend the methodology to a general nonlinear architecture which uses feedback. For these dynamic architectures, we describe how one can apply optimization techniques which make updates closer to the Newton direction than those of a steepest descent method, such as backpropagation.

Abstract: The method and system described herein use a biologically-based signal processing system for noise removal for signal extraction. A wavelet transform may be used in conjunction with a neural network to imitate a biological system. The neural network may be trained using ideal data derived from physical principles or noiseless signals to determine to remove noise from the signal.

Abstract: A system that reconstructs independent signals from degenerate mixtures estimates independent Auto Regressive (AR) processes from their sum. The system includes an identification system and an estimator. A mixture of two signals is inputted into the system and through the identifying and filtering processes, two estimates of the original signals are outputted. The identification system includes an ARMA identifier, a computation of autocovariance coefficients, an initializer and a gradient descent system. The estimator includes filtering.

Abstract: A neural network system is provided that models the system in a system model (12) with the output thereof providing a predicted output. This predicted output is modified or controlled by an output control (14). Input data is processed in a data preprocess step (10) to reconcile the data for input to the system model (12). Additionally, the error resulted from the reconciliation is input to an uncertainty model to predict the uncertainty in the predicted output. This is input to a decision processor (20) which is utilized to control the output control (14). The output control (14) is controlled to either vary the predicted output or to inhibit the predicted output whenever the output of the uncertainty model (18) exceeds a predetermined decision threshold, input by a decision threshold block (22).

Abstract: A method and apparatus of correcting for saturation in a current transformer, which outputs a current measurement, is provided. A switching algorithm receives a value of the current measurement from the current transformer and determines within which of three ranges the value falls. If the value falls in a first range, the current measurement is provided to a protective device such as a relay. If the value falls in a second range, the current measurement is provided to an artificial neural network that produces an output that accounts for saturation of the current transformer. If the value falls in a third range, the current measurement is provided to another artificial neural network that produces an output that accounts for saturation of the current transformer.

Abstract: The subject system provides reduced-dimension mapping of pattern data. Mapping is applied through conventional single-hidden-layer feed-forward neural network with non-linear neurons. According to one aspect of the present invention, the system functions to equalize and orthogonalize lower dimensional output signals by reducing the covariance matrix of the output signals to the form of a diagonal matrix or constant times the identity matrix. The present invention allows for visualization of large bodies of complex multidimensional data in a relatively “topologically correct” low-dimension approximation, to reduce randomness associated with other methods of similar purposes, and to keep the mapping computationally efficient at the same time.

Abstract: A predictive network is disclosed for operating in a runtime mode and in a training mode. The network includes a preprocessor (34') for preprocessing input data in accordance with parameters stored in a storage device (14') for output as preprocessed data to a delay device (36'). The delay device (36') provides a predetermined amount of delay as defined by predetermined delay settings in a storage device (18). The delayed data is input to a system model (26') which is operable in a training mode or a runtime mode. In the training mode, training data is stored in a data file (10) and retrieved therefrom for preprocessing and delay and then input to the system model (26'). Model parameters are learned and then stored in the storage device (22). During the training mode, the preprocess parameters are defined and stored in a storage device (14) in a particular sequence and delay settings are determined in the storage device (18).

Abstract: A method of predicting at least one future value of a time series of data using a neural network comprising the steps of:(I) inputting a plurality of values of the time series into the neural network;(ii) inputting information about a time into the neural network; and(iii) obtaining outputs from the neural network said outputs comprising predicted future value(s) of the time series.

Abstract: The present invention relates to a system and a method for signal classification. The system comprises a sensor array for receiving a series of input signals such as acoustic signals, pixel-based image signal (such as from infrared images detectors), light signals, temperature signals, etc., a wavelet transform module for transforming the input signals so that characteristics of the signals are represented in the form of wavelet transform coefficients and an array of hybrid neural networks for classifying the signals into multiple distinct categories and generating a classification output signal.

Abstract: A family of novel multi-layer discrete-time neural net controllers is presented for the control of an multi-input multi-output (MIMO) dynamical system. No learning phase is needed. The structure of the neural net (NN) controller is derived using a filtered error/passivity approach. For guaranteed stability, the upper bound on the constant learning rate parameter for the delta rule employed in standard back propagation is shown to decrease with the number of hidden-layer neurons so that learning must slow down. This major drawback is shown to be easily overcome by using a projection algorithm in each layer. The notion of persistency of excitation for multilayer NN is defined and explored. New on-line improved tuning algorithms for discrete-time systems are derived, which are similar to e-modification for the case of continuous-time systems, that include a modification to the learning rate parameter plus a correction term. These algorithms guarantee tracking as well as bounded NN weights.

Abstract: In order to realize a neural network for image processing by an inexpensive hardware arrangement, a neural network arranged in an image processing apparatus is constituted by an input layer having neurons for receiving information from picture elements in a 7.times.7 area including an interesting picture element in an image, an intermediate layer having one neuron connected to all the 49 neurons in the input layer and five groups of nine neurons, the nine neurons in each group being connected to nine neurons in the input layer, which receive information from picture elements in at least one of five 3.times.3 areas (1a to 1e), and an output layer having one neuron, which is connected to all the neurons in the intermediate layer and outputs information corresponding to the interesting picture element.

Abstract: Telecommunications processing is applied to a reference signal to generate a signal under test. A fidelity measure is generated characterizing the fidelity of the signal under test relative to the reference signal. A control signal is generated from the fidelity measure, where the control signal is used as a feedback signal to adjust the telecommunications processing. In one embodiment, the reference signal is a speech signal and the signal under test is a decoded speech signal generated by encoding, transmitting, and decoding the reference speech signal. The fidelity signal is an average mean opinion score (MOS) and the control signal is used to control the speech decoding processing. For example, the speech decoding processing may involve a speech decoder followed by a post filter, and the control signal is the cut-off frequency of the post filter.

Abstract: A recursive neurofilter comprising a recursive neural network (NN) is disclosed for processing an information process to estimate a signal process with respect to an estimation error criterion. The information process either consists of a measurement process, or if the signal and measurement processes are time-variant, consists of the measurement process as well as a time variance process, that describes the time-variant properties of the signal and measurement processes. The recursive neurofilter is synthesized from exemplary realizations of the signal and information processes. No assumptions such as Gaussian distribution, linear dynamics, additive noise, and Markov property are required. The synthesis is performed essentially through training recursive NNs. The training criterion is constructed to reflect the mentioned estimation error criterion with the exemplary realizations.

Abstract: Each neural element of a column-structured recurrent neural network generates an output from input data and recurrent data provided from a context layer of a corresponding column. One or more candidates for an estimated value is obtained, and an occurrence probability is computed using an internal state by solving an estimation equation determined by the internal state output from the neural network. A candidate having the highest occurrence probability is an estimated value for unknown data. Thus, the internal state of the recurrent neural network is explicitly associated with the estimated value for data, and a data change can be efficiently estimated.

Abstract: Source separation system for processing input signals (E.sub.i (t)) formed by instantaneous linear mixtures of primary signals (X.sub.j (t)) that result from sources (S1-Sn) and for producing at least one estimated primary signal (x.sub.k (t)) . The system includes separation apparatus (10) and characterization apparatus (15) which determine cumulants of the input signals for extracting estimated mixing coefficients (.alpha..sub.ij). The coefficients are transformed into separation coefficients (C.sub.ki, d.sub.ki) in the separation apparatus (10). The separation apparatus (10) may have a direct structure or a recursive structure.

Abstract: A likelihood of detecting a reflected signal characterized by phase discontinuities and background noise is enhanced by utilizing neural networks to identify coherency intervals. The received signal is processed into a predetermined format such as a digital time series. Neural networks perform different tests over arbitrary testing intervals to determine the likelihood of a phase discontinuity occurring in any such interval. An integration time generator subsequently uses this information to define a series of contiguous coherency intervals over the duration of the received signal. These coherency intervals are then used for piece-wise processing of the received signal by parallel quadrature receivers. The outputs are combined and processed for detecting the presence of the reflected signal.

Abstract: A nonrecurrent version of the Neural Network Binary Code Recognizer is disclosed. This Nonrecurrent Binary Code Recognizer, which decodes an input vector of n analog components into a decoded binary word of n bits, comprises an analog-to-digital converter, an inverter circuit, a digital summing circuit and a comparator circuit.

Abstract: In a method for tranining a neural network with the non-deterministic behavior of a technical system, weightings for the neurons of the neural network are set during the training using a cost function. The cost function evaluates a beneficial system behavior of the technical system to be modeled, and thereby intensifies or increases the weighting settings which contribute to the beneficial system behavior, and attenuates or minimizes weightings which produce a non-beneficial behavior. Arbitrary or random disturbances are generated by disturbing the manipulated variable with noise having a known noise distribution, these random disturbances significantly faciliating the mathematical processing of the weightings which are set, because the terms required for that purpose are simplified. The correct weighting setting for the neural network is thus found on the basis of a statistical method and the application of a cost function to the values emitted by the technical system or its model.