Information processing system using neural network learning function

• US 5,086,479 A
• Filed: 06/28/1990
• Issued: 02/04/1992
• Est. Priority Date: 06/30/1989
• Status: Expired due to Term

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Takenaga, Hiroshi, Tanaka, Norio, Asada, Kazuyoshi, Kitamura, Tadaaki, Takatoo, Masao, Kikuchi, Kuniyuki, Okuyama, Yoshiyuki
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US07/544,870
• Time in Patent Office: 586 Days
• Field of Search: 382/14, 382/15, 364/513
• US Class Current: 382/157
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning

Abstract

An information processing apparatus using a neural network learning function has, in one embodiment, a computer system and a pattern recognition apparatus associated with each other via a communication cable. The computer system includes a learning section having a first neural network and serves to adjust the weights of connection therein as a result of learning with a learning data signal supplied thereto from the pattern recognition apparatus via the communication cable. The pattern recognition apparatus includes an associative output section having a second neural network and receives data on the adjusted weights from the learning section via the communication cable to reconstruct the second neural network with the data on the adjusted weights. The pattern recognition apparatus with the associative output section having the reconstructed second neural network performs pattern recognition independently of the computer system with the communication cable being brought into an electrical isolation mode.

Claims

1. An information processing system using a neutral network learning function, comprising:

first operating means including learning data producing means for producing learning data for a neural network and associative output means having a first neural network for producing, after having been reconstructed on the basis of the learning data, output data responsive to given input data;

communication medium means; and

second operating means coupled to said first operating means by said communication medium means and including learning means supplied by said communication means with the learning data produced by said learning data producing means and having a second neural network for executing learning with the learning data to generate connection weight data to be supplied, via said communication medium means, from said second operating means to said first operating means so that reconstruction of said first neural network is effected with supplied connection weight data.

2. A pattern recognition system using a neural network learning function, comprising:

a pattern recognition apparatus including learning data producing means for producing learning data for a neural network and associative output means having a first neural network for producing, after having been reconstructed on the basis of the learning data, output data responsive to given input data;

communication medium means; and

operating means coupled to said pattern recognition apparatus by said communication medium means and including learning means, said learning means having a second neural network for executing learning by use of learning data produced by said learning data producing means to generate connection weight data, the learning data being supplied, via said communication medium means, from said pattern recognition apparatus to said operating means, and said connection weight data being supplied, via said communication medium means, from said operating means to said pattern recognition apparatus so that reconstruction of said first neural network is effected with supplied connection weight data.

3. A pattern recognition system according to claim 2, wherein said pattern recognition apparatus includes means for supplying structure information on the neural network, in addition to the learning data, to the operating means via said communication medium means for executing the learning by the second neural network.

4. A pattern recognition system according to claim 2, wherein said operating means includes the learning means and another associative output means.

5. An image recognition system using the neural network learning function, comprising:

an image recognition apparatus including learning data producing means for producing learning data for a neural network and associative output means having a first neural network for producing, after having been reconstructed on the basis of the learning data, output data responsive to given input data;

communication medium means; and

operating means coupled to said image recognition apparatus by said communication medium means and including learning means, said learning means having a second neural network for executing learning by use of learning data produced by said learning data producing means to generate connection weight data, the learning data being supplied via said communication medium means from said image recognition apparatus to said operating means, and said connection weight data being supplied via said communication medium means from said operating means to said image recognition apparatus so that reconstruction of said first neural network is effected with the supplied connection weight data.

6. An image recognition system according to claim 5, wherein the learning data producing means of the image recognition apparatus includes means for producing the learning data on the basis of image data of an object for recognition or features extracted from the object for recognition or both.

7. An image recognition system using a neural network according to claim 5, further comprising first display means for displaying a configuration of the neural network used for a selected one of the learning and the associative output.

8. An image recognition system according to claim 7, further comprising enlarged display means for displaying in enlarged form a predetermined portion of the configuration of a neural network displayed by the first display means, and correcting means for executing at least a selected one of the correction of the threshold of neurons for the neural network displayed in enlarged form, the correction of the weights between neurons and the disconnection of neurons.

9. An image recognition system according to claim 7, wherein the first display means includes a selected one of means for displaying a selected one of the color and brightness changed in accordance with the degree of excitement of neurons and blink display means for displaying by blinking neurons high in the degree of excitement.

10. An image recognition system according to claim 5, wherein said image recognition apparatus includes a user application having input data producing means for producing input data and discriminating means for making discrimination by use of the result of the associative output from the associative output means, and interface means permitting the user application to be connected with the learning data producing means and the associative output means.

11. An image recognition system according to claim 10, wherein said interface means is common table means.

12. An image recognition system according to claim 10, further comprising:

image pick-up means for picking up an image of an automobile including a registration number plate and supplied with the particular image; and

pre-processing means for extracting the registration number plate from the image supplied from the image pick-up means and selecting the registration number of the number plate thus extracted and feature extracting means for extracting the features from the registration number selected by the pre-processing means, said pre-processing means and said feature extracting means being included in the user application;

wherein the learning means learns from the features extracted from the feature extracting means, and the associative output means produces an associative output by use of the features extracted.

13. An image recognition system according to claim 10, further comprising:

image pick-up means for picking up an image of an automobile including a registration number plate and supplied with the particular image; and

pre-processing means in the user application for extracting the registration number plate from the image supplied from the image pick-up means and selecting the registration number from the number plate thus extracted;

wherein the learning means learns from the data of the registration number selected by the preprocessing means, and the associative output means produces an associative output by use of the data of the registration number thus selected.

14. A pattern recognition apparatus using a neural network learning function, comprising:

an input signal source section for supplying at least one of a learning input signal and a pattern input signal representative of a pattern to be recognized;

a communication medium;

a learning section associated with said input signal source section via said communication medium for receiving data concerning the learning input signal, said learning section including first neural network means and weight adjusting means coupled to said first neural network means for adjusting weights of connections in said first neural network means so as to minimize the difference between an output of said first neural network means in response to received data concerning the learning input signal and a teacher'"'"'s data signal;

an associative output section including second neural network means and associated with said learning section via said communication medium for receiving therefrom the adjusted weight data to for incorporation thereof into said second neural network means, said associative output section being coupled to said input signal source section for receiving therefrom the pattern input signal so that said associative output section having the adjusted weight data incorporated thereinto produces an associative output signal in response to the pattern input signal, said communication medium associating said learning section with said associative output section being arranged such that they are electrically isolated from each other when said associative output section is in operation for producing the associative output; and

a discriminator section coupled to said associative output section for determining whether the associative output signal represents the pattern input signal or not.

15. A pattern recognition apparatus according to claim 14, further comprising a learning data producing section coupled to said input signal source section for receiving the learning input signal therefrom and for producing a learning data signal by combining the learning input signal with the teacher'"'"'s data signal, said learning section being associated with said input signal source section by way of said learning data producing section via said communication medium for receiving the learning data signal.

16. A pattern recognition apparatus according to claim 14, further comprising means for supplying a teacher'"'"'s data signal to said learning section.

17. An information processing system comprising:

a communication line;

a function defining apparatus connected through a first communication interface with said communication line, said function defining apparatus including a learning section having first neural network means and weight adjusting means coupled to said first neural network means for adjusting weights of connections in said first neural network means so as to minimize the difference between an output of said first neural network means in response to a learning input signal and a teacher'"'"'s data means in response to a learning input signal and a teacher'"'"'s data signal; and

a plurality of information processing apparatuses each connected with said communication line through a second communication interface, each of said information processing apparatuses including a learning data producing section for producing a learning data signal containing the learning input signal and the teacher'"'"'s data signal, the learning data signal being transferred by said second communication interface to said learning section via said communication line and said first communication interface, and each of said information processing apparatuses further including an associative output section having second neural network means and being operable to receive the data on the adjusted weights to incorporate them into said second neural network means, the data on the adjusted weights being transferred by said first communication interface to said associative output section via said communication line and said second communication interface, said associative output section being coupled to an input signal source for receiving therefrom a pattern input signal representative of a pattern to be recognized so that said associative output section having the adjusted weight data incorporated thereinto produces an associative output in response to the pattern input signal.