Method and apparatus for checking input-output characteristic of neural network

• US 5,333,238 A
• Filed: 08/28/1992
• Issued: 07/26/1994
• Est. Priority Date: 08/29/1991
• Status: Expired due to Fees

• Current Assignee: Toyoda Koki Kabushiki Kaisha, Yukinori Kakazu
• Original Assignee: Toyoda Koki Kabushiki Kaisha, Yukinori Kakazu
• Inventors: Kato, Tomonari, Hattori, Kazuya, Kakazu, Yukinori, Yoneda, Takao
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US07/936,635
• Time in Patent Office: 697 Days
• Field of Search: 395/21, 395/23, 395/22, 395/24, 395/128, 364/550, 364/553
• US Class Current: 706/44
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/10 Interfaces, programming languages or software development kits, e.g. for simulating neural networks

Abstract

An apparatus for checking the input-output characteristic of a neural network which has an input layer, an intermediate layer and an output layer. Plural nodes of the input layer are related to plural nodes of the intermediate layer with plural connection weights while the plural nodes of the intermediate layer are also related to plural nodes of the output layer with plural connection weights. One of the nodes of the input layer is selected as a variable input element while the rest of the nodes are regarded as fixed input elements. A variable data input to the variable input element is varied within a predetermined variation range, while data input to the fixed input elements are fixed, so as to detect how the components of output data from the output layer change. The detected changes of the components of the output data are displayed as the input-output characteristic of the neural network.

Claims

1. A method of checking an input-output characteristic of a neural network comprising an input layer composed of plural input nodes and an output layer composed of at least one output node, said method comprising the steps of:

selecting one of said input nodes;

applying a variable data to said selected input node and applying constant data to input nodes other than said selected node;

operating said neural network to obtain output data from said output layer of said neural network;

varying said variable data within a predetermined range;

operating said neural network to obtain output data from said output layer of said neural network; and

displaying a change of said output data.

6. A method for checking an input-output characteristic of a neural network according to claim 1, further comprising:

setting said variable data to a lower limit of said predetermined range;

operating said neural network using said variable data set to said lower limit;

incrementing said variable data by a variation amount;

operating said neural network using said variable data incremented by said variation amount; and

repeating said steps of incrementing and said operating said neural network using variable data incremented by said variation amount until said variable data reaches an upper limit of said predetermined range.

7. A method for checking an input-output characteristic of a neural network according to claim 6, further comprising:

determining a usable range of said variable data based upon said displaying of said output data.

8. A method for checking an input-output characteristic of a neural network according to claim 1, further comprising:

determining a usable range of said variable data based upon said displaying of said output data.

9. A method for checking an input-output characteristic of a neural network according to claim 1, further comprising:

selecting a second node of said input layer;

applying second variable data to said second input node and applying said constant data to input nodes other than said selected node;

operating said neural network to obtain output data using said second variable data;

varying said second variable data within a second predetermined range; and

operating said neural network using said varied second variable data.

10. A method for checking an input-output characteristic of a neural network according to claim 1, further comprising:

repeating said selecting, applying, operating and varying steps for all other input nodes of said neural network.

2. An apparatus for checking an input-output characteristic of a neural network comprising an input layer having plural input nodes and an output layer having one output node, said apparatus comprising:

means for selecting one of said input nodes;

means for applying a variable data to said selected input node and applying constant data to input nodes other than said selected node;

means for operating said neural network to obtain output data from said output layer of said neural network;

means for varying said variable data within a predetermined range; and

means for displaying a change of said output data.

3. An apparatus for checking an input-output characteristic of a neural network according to claim 2, wherein said means for varying said data comprises:

means for inputting a variation range within which said data is to be changed;

means for inputting a variation amount;

means for initially setting said variable data to a lower limit of said variation range; and

means for repeatedly increasing said variable data by said variation amount until said variable data reaches an upper limit of said variation range.

4. An apparatus for checking an input-output characteristic of a neural network according to claim 3, wherein said apparatus further comprises means for inputting fixed data to said input nodes other than said selected node.

5. An apparatus for checking an input-output characteristic of a neural network according to claim 3, wherein said neural network further comprises an intermediate layer composed of having plural nodes.

11. A method of checking an input-output characteristic of a neural network comprising an input layer composed of plural input nodes and an output layer composed of at least one output node, said method comprising the steps of:

selecting one of said input nodes of said neural network;

applying variable data to said one input node while applying constant data to input nodes other than said one input node and operating said neural network to obtain first output data from said output node;

varying said variable data within a predetermined range to obtain second variable data;

applying said second variable data to said one input node while applying said constant data to said input nodes other than said one input node and operating said neural network to obtain second output data from said output node; and

displaying said first and second output data.

12. A method for checking an input-output characteristic of a neural network according to claim 11, further comprising:

setting aid variable data to a lower limit of said predetermined range;

operating said neural network using said variable data set to said lower limit;

incrementing said variable data by a variation amount;

operating said neural network using said variable data incremented by said variation amount; and

repeating said incrementing step and said operating said neural network using variable data incremented by said variation amount step until said variable data reaches an upper limit of said predetermined range.

13. A method for checking an input-output characteristic of a neural network according to claim 12, further comprising:

determining a usable range of said variable data based upon said output data.

14. A method for checking an input-output characteristic of a neural network according to claim 11, further comprising:

determining a usable range of said variable data based upon said displaying of said output data.