Method for processing data using a neural network having a number of layers equal to an abstraction degree of the pattern to be processed

• US 5,553,196 A
• Filed: 06/05/1995
• Issued: 09/03/1996
• Est. Priority Date: 04/05/1989
• Status: Expired due to Fees

• Current Assignee: Sharp Corporation (Hon Hai Precision Industry Co., Ltd.), YOZAN, Inc.
• Original Assignee: Sharp Corporation (Hon Hai Precision Industry Co., Ltd.), YOZAN, Inc.
• Inventors: Matsumoto, Koji, Yamamoto, Makoto, Takatori, Sunao, Kumagai, Ryohei
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/461,538
• Time in Patent Office: 456 Days
• Field of Search: 395/22, 395/23, 395/24, 395/26, 395/27
• US Class Current: 706/31
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06T 7/00 Image analysis
• CPC Class Codes: G06V 10/46 Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features colour feature extraction G06V10/56

Abstract

There is provided a layer construction of neural layers according to the abstraction degree of data to be processed, and data is inputted to a neural layer corresponding to its abstraction degree.

Claims

1. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network, a number of processing layers within said neural network corresponding to a number of different abstraction degrees associated with said input data;

generating an output from said neural network corresponding to said input data using said neural network; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.

2. A method for processing data according to the method of claim 1, wherein the number of processing layers within said neural network is increased with an increase in a number of different abstraction degrees associated with said input data.

3. A method for processing data according to the method of claim 1, wherein the number of processing layers within said neural network is increased proportionately with an increase in a number of different abstraction degrees associated with said input data.

4. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network having a number of processing layers equal to a number of different abstraction degrees associated with said input data;

generating an output from said neural network corresponding to said input data; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.

5. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network having a number of processing layers equal to a number of different abstraction degrees associated with said input data, whereby input data having a same abstraction degree are input to one of said processing layers;

generating an output from said neural network corresponding to said input data; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.

6. A method of processing as recited in claim 5, whereby at least one group of input data characterized with a same abstraction degree is input to each processing layer in said neural network.

7. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network having a number of processing layers equal to a number of different abstraction degrees associated with said input data, where all input data having a lowest abstraction degree are input to a first processing layer, and where input data respectively input to each subsequent processing layer in said network increases sequentially in abstraction degree;

generating an output from said neural network corresponding to said input data; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.

8. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network, a number of processing layers within said neural network being related to a number of different abstraction degrees associated with said input data;

generating an output from said neural network corresponding to said input data using said neural network; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.

9. A method for processing data including the steps of:

inputting a plurality of input data to be processed, each input data having an associated abstraction degree;

forming a neural network, wherein a relationship exists between a number of processing layers within said neural network and a number of different abstraction degrees associated with said input data;

generating an output from said neural network corresponding to said input data using said neural network; and

determining a pattern of features of said input data corresponding to said output generated from said neural network.