Intelligent control with hierarchical stacked neural networks

• US 9,053,431 B1
• Filed: 07/02/2014
• Issued: 06/09/2015
• Est. Priority Date: 10/26/2010
• Status: Active Grant

• Current Assignee: Michael Lamport Commons
• Original Assignee: Michael Lamport Commons
• Inventors: Commons, Michael Lamport
• Primary Examiner(s): CHANG, LI WU
• Application Number:
  US14/322,147
• Time in Patent Office: 342 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: G10L 15/16 using artificial neural networks

Abstract

A system and method of detecting an aberrant message is provided. An ordered set of words within the message is detected. The set of words found within the message is linked to a corresponding set of expected words, the set of expected words having semantic attributes. A set of grammatical structures represented in the message is detected, based on the ordered set of words and the semantic attributes of the corresponding set of expected words. A cognitive noise vector comprising a quantitative measure of a deviation between grammatical structures represented in the message and an expected measure of grammatical structures for a message of the type is then determined. The cognitive noise vector may be processed by higher levels of the neural network and/or an external processor.

Claims

1. A system configured to analyzing at least one data pattern comprising:

an input configured to receive at least one data pattern;

at least one hierarchical neural network, having a plurality of hierarchical layers, each respective hierarchical layer being configured to receive a respective input and to produce a non-arbitrary organization of actions in dependence on the respective input and a respective layer training, the at least one hierarchical neural network comprising;

a first layer configured to produce a non-arbitrary organization of actions which identifies at least one data object from a plurality of data objects, based at least the first layer training to identify a plurality of different data objects, and the at least one data pattern, and to produce a noise vector output, distinct from the non-arbitrary organization of actions of the first layer, representing a deviance of at least a portion of the at least one data pattern from a prototype of the data object identified,

a second layer, configured to receive the respective non-arbitrary organization of actions from the first layer identifying the object as the respective input, based on the non-arbitrary organization of actions from the first layer, to ascertain a type of the identified data object from a plurality of different types of each of the plurality of different data objects;

wherein the first layer further produces a noise vector output, distinct from the non-arbitrary organization of actions of the first layer, representing a deviance of at least a portion of the at least one data pattern from a prototype of the data object identified, and

a processor configured to at least one of;

determine a confidence of data object identification, and determine that the data pattern comprises a data object not properly identified by the first layer.

2. The system according to claim 1, wherein the data pattern comprises at least one image.

3. The system of claim 2, wherein the at least one image comprises a video.

4. The system of claim 1, wherein the second layer is further configured to produce a second noise vector output, distinct from the non-arbitrary organization of actions of the second layer, representing a deviance of the identified object from a prototype of the ascertained type of object, and the at least one hierarchical neural network further comprises a third layer configured to receive the non-arbitrary organization of actions from the second layer, and to further process the information from the at least one data pattern propagated through the first layer and the second layer in dependence on respective training of the third layer.

5. The system of claim 1, wherein each of the first layer and the second layer of the hierarchical neural network comprises a plurality of neurons.

6. The system according to claim 1, wherein the data pattern comprises semantic information and the respective layer training is adapted to perform a semantic analysis according to a predefined grammar.

7. A method for analyzing at least one data pattern with at least one hierarchical neural network, having a plurality of layers, each layer being configured to receive a respective input and to produce a non-arbitrary organization of actions in dependence on the respective input and a respective layer training, the at least one hierarchical neural network comprising:

a first layer configured to identify at least one data object in at least one data pattern presented at the respective input, from a plurality of different data objects, based on at least the first layer training and the at least one data pattern, and to produce a noise vector output, distinct from the non-arbitrary organization of actions of the first layer, representing a deviance of at least a portion of the at least one data pattern from a prototype of the data object identified,

a second layer, configured to receive the respective non-arbitrary organization of actions from the first layer identifying the data object as the respective input, and, based on the respective layer training of the second layer and the respective non-arbitrary organization of actions from the first layer, to ascertain a type of the identified data object from a plurality of different types of each of the plurality of different data objects; and

a processor configured to at least one of determine a confidence of a data object identification, and determine that the data pattern comprises a data object not properly identified by the first layer,

the method comprising;

receiving the at least one data pattern at a lowest hierarchical layer;

processing the at least one data patter to identify the data object;

producing the noise vector representing the deviance of at least a portion of the at least one data pattern from a prototype of the data object identified;

processing the non-arbitrary organization of actions from the first layer with the second layer, to ascertain the type of identified object; and

determining with the processor at least one of the confidence of a data object identification, and that the data pattern comprises a data object not properly identified by the first layer.

8. The method according to claim 7, wherein the data pattern comprises image data.

9. The method according to claim 7, wherein the second layer further produces a second noise vector output, distinct from the non-arbitrary organization of actions of the second layer, representing a deviance of the identified object from a prototype of the ascertained type of object, and the at least one hierarchical neural network further comprises a third layer receiving the non-arbitrary organization of actions from the second layer, and further processing the information from the at least one data pattern propagated through the first later and the second layer in dependence on respective training of the third layer.

10. The method according to claim 7, wherein each of the first layer and the second layer of the hierarchical neural network comprises a plurality of neurons of an artificial neural network.

11. The method according to claim 7, wherein the data pattern comprises semantic information.

12. The method according to claim 11, wherein the semantic information is organized according to a predefined grammar.

13. The method according to claim 7, further comprising employing at least the non-arbitrary organization of actions of the second layer to define an input to a text input Internet search engine.

14. The method according to claim 7, further comprising employing at least the noise vector to define an input to the text input Internet search engine.

15. The method according to claim 7, wherein the data pattern comprises a semantic message having a message type, further comprising:

detecting an ordered set of words within the semantic message;

linking the set of words detected within the semantic message to a corresponding set of expected words, the set of expected words having semantic attributes, the set of expected words being based on at least the non-arbitrary organization of actions of the second layer.

16. The method according to claim 15, further comprising detecting a set of grammatical structures represented in the message, based on the ordered set of words and the semantic attributes of the corresponding set of expected words.

17. The method according to claim 15, further comprising:

searching a database comprising semantic attributes for at last one record corresponding to semantic attributes of the semantic message; and

outputting an indication of the at least one record comprising semantic attributes corresponding to semantic attributes of the semantic message.

18. The method according claim 15, wherein the semantic attributes correspond to proper spelling, grammar, and word use in the spoken or written language.

19. The method according to claim 15, further comprising generating a signal dependent on a degree to which the set of grammatical structures represented in the semantic message are consistent with the semantic attributes of the set of expected words.

20. A method for analyzing at least one data pattern, with at least two layer of an interconnected hierarchical neural network, each layer receiving a respective input and producing a non-arbitrary organization of actions in dependence on a respective layer input and a respective layer training, comprising:

providing a first layer of the interconnected hierarchical neural network, which identifies a data object represented in a data pattern presented at the respective first layer input from a plurality of different data objects which form a basis for the first layer training, and produce a noise vector output, distinct from the non-arbitrary organization of actions of the first layer, representing a deviance of at least a portion of the data pattern from a prototype of the data object identified;

providing a second layer of the interconnected hierarchical neural network, which receives the respective non-arbitrary organization of actions from the first layer as the respective second layer input, and classifies the identified data object from a plurality of different types of the identified data object from a plurality of different data object types which form a basis for the second layer training; and

determining a confidence of the identification of the data object by the first layer based on at least the non-arbitrary organization of actions of the first layer and the noise vector output.