Neural network system adapted for non-linear processing

• US 5,402,519 A
• Filed: 11/25/1991
• Issued: 03/28/1995
• Est. Priority Date: 11/26/1990
• Status: Expired due to Term

• Current Assignee: Hitachi Engineering Co.,Ltd (Hitachi, Ltd.), Hitachi, Ltd.
• Original Assignee: Hitachi Engineering Co.,Ltd (Hitachi, Ltd.), Hitachi, Ltd.
• Inventors: Koide, Minoru, Nakamura, Kenichi, Inoue, Haruki, Abe, Noboru, Oshima, Keiji, Yahiro, Masakazu
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US07/797,534
• Time in Patent Office: 1,219 Days
• Field of Search: 395/11, 395/3, 395/51, 395/61, 395/900, 395/22
• US Class Current: 706/16
• CPC Class Codes: G06N 3/043 based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 7/046 Implementation by means of a neural network neural networks using fuzzy logic G06N3/043
• CPC Class Codes: Y10S 706/90 Fuzzy logic

Abstract

A neural network system includes a qualitative evaluation section, a neural network section, a quantifying section and a display section. The qualitative evaluation section qualitatively analyzes an unknown data supplied thereto, and normalizes the result of analysis within a predetermined range. The neural network section having a neural network with plural neurons computes the network output data from the normalized unknown data produced by the qualitative evaluation section. Each neuron is connected to plural other neurons through synapses, each of which is assigned an individual weight coefficient. Each neuron is adapted to output an output function value assigned thereto associated with the total sum of the products of the output from the neurons connected thereto and the synapse weight coefficient. The quantifying section quantifies the network output data to produce desired data. The desired data thus produced is displayed on the display section.

Claims

1. A neural network system comprising:

qualitative evaluation means for qualitatively analyzing unknown data supplied thereto and normalizing the qualitatively analyzed unknown data within a predetermined range;

computing means, including a neural network having a plurality of neurons, for computing desired data from the normalized unknown data from said first qualitative evaluation means, each neuron being connected to other neurons through synapses, each synapse being given a weight, each neuron outputting a value of an output function given to the neuron in accordance with a total sum of product values of output function values from said connected neurons and said synapse weights,

wherein said computing means further includes;

means for storing a causal network representing qualitative causal relation between the unknown data and the desired data, and

network defining means for defining said neural network by connecting neurons through synapses in accordance with the causal network; and

quantitative evaluation means for quantitatively analyzing desired data received from said computing means and outputting the quantitatively analyzed desired data.

2. A neural network system according to claim 1, wherein said computing means further comprises:

removing means for removing computation on the synapses whose weights are smaller than a predetermined value and a neuron, weights of the synapses connected to which are smaller than said predetermined value, from the computation of the desired data based on the analyzed unknown data.

3. A neural network system according to claim 1, wherein said computing means further comprises:

table storage means for storing a product value table which stores product values and a function value table which stores output function values; and

means for determining a product value for each output function value of neurons connected to said each neuron with reference to said product value table, determining a total sum of the product values, and determining an output function value of neurons with reference to the function value table according to the total sum.

4. A neural network system according to claim 3, wherein

the neurons of said neural network are layered, and

said computing means further includes means for causing said neural network to synchronously perform processing for referring to said product value table, processing for determining the total sum, and processing for referring to said function value table, in units of layered neurons.

5. A neural network system according to claim 3, wherein said computing means further comprises:

removing means for removing computation on the synapse whose weight is smaller than a predetermined value and a neuron, weight of the synapse connected to which are smaller than the predetermined value, from computation of the desired data from the unknown data.

6. A neural network system comprising:

first evaluation means for qualitatively analyzing unknown data supplied thereto;

computing means, including a neural network having a plurality of neurons, for computing desired data from the qualitatively analyzed unknown data output from said first evaluation means, each neuron being connected through synapses to other neurons, each synapse being given a weight, each neuron outputting an output function value in accordance with a total sum of products of output function values from the connected neurons and synapse weights,

wherein said computing means further includes;

means for storing a network of qualitative causal relation between unknown and desired data, and

network definition means for defining the neural network by connecting neurons through synapses in accordance with the qualitative causal network; and

second evaluation means for quantitatively analyzing desired data received from said computing means and outputting the result of analysis.

7. A neural network system according to claim 6, wherein said computing means further comprises:

removing means for removing computation on the synapse whose weight is smaller than a predetermined value, and a neuron, weights of the synapses connected to which are smaller than said predetermined value, from computation of the desired data from the analyzed unknown data.

8. A neural network system according to claim 6, wherein said computing means further comprises:

table storage means for storing a product value table which stores product values and a function value table which stores output function values; and

means for determining the product value for each output function value of neurons referring to the product value table in response to the supply of the output function values of the neurons connected to sail each neuron, determining the total sum of the product values, and determining an output function value with reference to the function value table associated with the total sum thereby to determine the output function value of the particular neurons.

9. A neural network system according to claim 8, wherein

the neurons of the neural network are tiered, and

said computing means further includes means for determining the output function value of the neurons in a given tier with reference to the function value table subsequent to the product value table in synchronism with the particular neurons.

10. A neural network system according to claim 8, wherein said computing means further comprises:

removing means for removing synapses of the neural network means whose weight coefficients are smaller than a predetermined value, and also a neuron and synapses connected thereto whose weight coefficients are all smaller than a predetermined value, from the computation of desired data based on unknown data.

11. A neural network system comprising:

fuzzy means for fuzzy conversion of an unknown data supplied thereto;

neural network means including a neural network having a plurality of neurons for computing the network output data from the unknown data fuzzy-converted by said fuzzy means, each neuron being connected through synapses to other neurons, each synapse being given a weight coefficient, each neuron being adapted to output an output function value assigned thereto associated with the total sum of the products of the output from the neurons connected and the synapse weight coefficient,

wherein said neural network means further includes;

means for storing a network of qualitative causal relation between unknown and desired data, and

network definition means for defining the neural network by connecting neurons through synapses in accordance with the qualitative causal network; and

defuzzy means for defuzzy-converting output data of said neural network means and outputting a desired data.

12. A neural network system according to claim 11, wherein said neural network means further comprises:

removing means for removing synapses in the neural network whose weight coefficients are smaller than a predetermined value and a neuron, weights of the synapses connected thereto which are all smaller than a predetermined value, from the computation of the network output data based on unknown data.

13. A neural network system according to claim 11, wherein said neural network means further comprises:

table storage means for storing a product value table which stores product values and a function value table which stores output function values, and

means for determining the product value for each output function value of neurons with reference to the product value table in response to the supply thereto of the output function value of neurons connected to each neuron, determining the total sum of the product values, and determining an output function value with reference to the function value table associated with the total sum to thereby determine the output function value of the particular neurons.

14. A neural network system according to claim 13, wherein

the neurons of said neural network are layered, and

said neural network means further includes means for determining the output function value of the neurons in a given layer with reference to the function value table subsequent to the product value table in synchronism with the particular neurons.

15. A neural network system according to claim 13, wherein said neural network means further comprises:

removing means for removing the synapses of the neural network whose weight coefficients are smaller than a predetermined value, and also a neuron and synapses connected thereto whose weight coefficients are all smaller than a predetermined value, from the computation of desired data based on unknown data.