Fujitsu Corporation

Other

Asakawa, Kazuo

Kazuo Asakawa

Takashi Kimoto

Akira Kawamura

Nobuo Watanabe

Hideki Kato

Ryusuke Masuoka

Hideki Yoshizawa

Hiroki Iciki

Chikara Tsuchiya

Katsuya Ishikawa

Nobuhiko Onda

Tadashi Akita

Takashi Kawasaki

Fumio Tabata

Hiromu Iwamoto

Toru Kamada

Yoshihide Sugiura

Jun'ichi Tanahashi

Toshiharu Matsuda

Hiroyuki Tsuzuki

Hiroyuki Okada

Hitoshi Komoriya

Yuri Owada

Ryushuke Masuoka

Shigenori Matsuoka

Fumiaki Akiya

Hideichi Endo

Kazuo Kawabata

Yasuyuki Oishi

Shigemi Nagata

Takeshi Takano

Eisuke Fukuda

Toshimasa Miyazaki

Yoshinori Yaginuma

Hideichi Endoh

The present invention relates to a data processing system in a hierarchical network configuration for executing applicable data processes in a comprehensible and executable form. An object of the present invention is to allow data processing capabilities to be established with high precision in a short time based on a fuzzy-neuro-integrated concept. A fuzzy model is generated by a data processing system in the form of membership functions and fuzzy rules as technical information relating to a control target. According to this fuzzy model, a weight value of the connection between neurons is set and a pre-wired neural network is established. Then, the data of the control target are learned by the neural network. The connection state and a weight value of the neural network after the learning enable tuning of the fuzzy model.

Neuro-fuzzy-integrated data processing system

A learning system is used in a data processing apparatus for learning an input pattern by obtaining an internal-state value necessary for realizing a desired data conversion by performing a pattern conversion defined by the internal-state value and calculating an output pattern corresponding to the input pattern. The learning system comprises a pattern presenting unit for presenting an input pattern group of the subject to be learned for pattern conversion, dividing the input pattern group of the subject to be learned into at least two sets, selecting one of the divided sets, presenting the input pattern group of the selected set to a pattern conversion unit and presenting an input pattern group belonging to all the sets presented up to the current point when the internal-state value to be converged is obtained in accordance with the presentation of the selected set, and an error value calculating unit for calculating an error value representing a magnitude of a non-consistency between an output pattern group outputted in accordance with the presentation and a teacher pattern group representing a pattern to be obtained by the output pattern group.

Learning system for a data processing apparatus

A parallel data processing system comprises a plurality of data processing units each having at least one input and storing data of a matrix and a plurality of trays each having a first input and an output and for storing data of a vector, each of all or part of said trays having a second output connected to said first input of a respective one of said data processing units, and said trays being connected in cascade to form a shift register for performing data transfer between corresponding ones of the trays and the data processing units and data processing in the data processing units synchronously, thereby performing an operation of a matrix vector product or a neuron computer operation on analog signals.

Ring systolic array system for synchronously performing matrix/neuron

An adaptive equalizer detects characteristics of a transmission line in radio and cable communications and equalizes received signals by using the detected results. Equalization is for not only linear distortion but also nonlinear distortion in accordance with variation of characteristics of the transmission line. The adaptive equalizer includes a characteristic detection unit 11 for detecting real part and imaginary parts of a transfer function by using an output signal corresponding to an input signal of an object to be equalized to compensate for deterioration, for example a transmission line; and an equalizing unit 12 for equalizing an output signal of the object to be equalized 10 in accordance with an unknown input signal by using the detected result of the characteristic detection unit 11.

Adaptive equalizers

A learning system is used in a data processing apparatus for learning an input pattern by obtaining an internal-state value necessary for realizing a desired data conversion by performing a pattern conversion defined by the internal-state value and calculating an output pattern corresponding to the input pattern. The learning system comprises a pattern presenting unit for presenting an input pattern group of the subject to be learned for pattern conversion, dividing the input pattern group of the subject to be learned into at least two sets, selecting one of the divided sets, presenting the input pattern group of the selected set to a pattern conversion unit and presenting an input pattern group belonging to all the sets presented up to the current point when the internal-state value to be converged is obtained in accordance with the presentation of the selected set, and an error value calculating unit for calculating an error value representing a magnitude of a non-consistency between an output pattern group outputted in accordance with the presentation and a teacher pattern group representing a pattern to be obtained by the said output pattern group.

A learning process system is provided for a neural network. The neural network is a layered network comprising an input layer, an intermediate layer and an output layer formed of basic units. In the basic units, a plurality of inputs is multiplied by a weight signal and the products are accumulated, thereby supplying the sum of products. An output signal is obtained using a threshold value function in response to the sum of products. An error signal is generated by an error circuit in response to a difference between the output signal obtained from the output layer and a teacher signal. A weight updating signal is determined in a weight learning circuit by obtaining a weight value in which the sum of the error values falls within an allowable range. Thus, the learning is performed in the layered neural network through use of a back propagation method. Through such learning in the layered neural network, an updating quantity to be obtained in the present weight updating cycle is determined in response to a once delayed weight updating quantity signal in a previous weight updating cycle and a twice delayed weight updating quantity obtained at a twice-previous weight updating cycle prior to the previous weight updating cycle.

Learning process system for use with a neural network structure data

A waveform equalizer for equalizing a distorted signal, contains a sampling unit, a time series generating unit, and an equalization neural network unit. The sampling unit samples the level of a distorted signal at a predetermined rate. The time series generating unit serially receives the sampled level and outputs in parallel a predetermined number of the levels which have been last received. The equalization neural network unit receives the outputs of the time series generating unit, and generates an equalized signal of the distorted signal based on the outputs of the time series generating unit using a set of equalization network weights which are preset therein. The waveform equalizer may further contain a distortion characteristic detecting unit, an equalization network weight holding unit, and a selector unit. The distortion characteristic detecting unit detects a distortion characteristic of the distorted signal. The equalization network weight holding unit holds a plurality of sets of equalization network weights each for being set in the equalization neural network unit. The selector unit selects one of the plurality of sets of equalization network weights according to the distortion characteristic which is detected in the distortion characteristic detecting unit, and supplies the selected set in the equalization neural network unit to set the selected set therein.

Waveform equalizer using a neural network

An input analog data is applied to a plurality of neuron units in a time division manner. The analog input data is multiplied by digital weight data which can be changed in accordance with the data of the interconnection between units. The products of the time division analog input data and the digital weight data are added in an integrator. While the present sum of the products is output, the previous sum of the products is output simultaneously with the present data, thereby providing outputs in a pipe-line manner. When the output of the first neuron is produced, the second neuron in the same layer produces an output such that the output of the first layer is produced on the output analog bus in a time division manner. This analog neuron unit constitutes an intermediate layer and an output layer. One layer of neuron units can be repeatedly used by feeding back the output of one layer to the input of another layer, then the neuron system operates as a layered structure.

Neuron architecture

An error absorbing system for absorbing errors through a weight correction is provided in a neuron computer for receiving an analog input signal through a first analog bus in a time divisional manner, performing a sum-of-the-products operation, and outputting an analog output signal to a second analog bus. The error absorbing system includes a dummy node for producing a fixed voltage to an analog bus in a test mode. The dummy node is connected to the analog bus of the neural network. An error measuring unit compulsorily inputs 0 volts to the first analog bus through the dummy node in a first state of a test mode and detects an offset voltage produced in an analog neuron processor through the second analog bus. A weight correcting unit, in a second state of the test mode, determines a temporary weight between the dummy node and the neuron processor. The temporary weight is multiplied by the fixed voltage produced by the dummy node, based on an offset voltage of respective neuron processors. The weight correcting unit calculates a correct weight using a gain based on the detection output voltage output from the second analog bus. A weight memory stores the weight corrected by the weight correcting unit.

Error absorbing system in a neuron computer

A supporting device comprising a first member which supports a springy means for resiliently supporting a second member. The supporting device comprises a detection means for detecting displacement of the springy means with respect to the first member in accordance with the movement of the second member and a biasing means which applies a force to the springy means from the same direction as or from a direction opposite to that of the displacement for changing the elasticity of the springy means in accordance with the movement of the second member.

Supporting device

Neurocomputer with analog signal bus

In a force controlling system, a movable member fixed to an elastic member applies a force in correspondence with a force information signal to an object. At this time, an additional force is applied via a driver to the movable member by the positive feedback of the displacement of the elastic member to the driver, so that the reactive force due to the displacement of the elastic member is contradicted by the additional force. As a result, the movable member applies only a force in correspondence to the force information signal to the object.

Force controlling system

A supporting device including a first member which supports a springy means for resiliently supporting a second member. The supporting device includes a detection for detecting displacements of the springy with respect to the first member in accordance with the movement of the second member and a biasing which applies a force to the springy from the same direction as or from a direction opposite to that of the displacement for changing the elasticity of the springy in accordance with the movement of the second member.

Growing community of inventors

Kawasaki, Japan

Kazuo Asakawa