Back to results
US 8,775,341 B1Granted
Intelligent control with hierarchical stacked neural networks
OwnerMichael Lamport Commonsinventor
Life of this patent
Estimating the term…
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.
Claims20
A method of processing information, comprising:
- receiving at least one message;
- processing the at least one message with an artificial neural network based processor, trained to respond to received information and produce at least one set of outputs which represent intelligent discriminations of information in at least one respective stage of cognitive development in a non-arbitrary organization of actions based on an architecture of the artificial neural network based processor and a training of the artificial neural network based processor;
- representing as a noise vector at least one data pattern in the at least one received message which is incompletely represented as the intelligent discriminations of information in the at least one set of outputs representing the non-arbitrary organization of actions;
- analyzing at least the noise vector distinctly from the artificial neural network based processor to extract characteristics of the at least one message distinct from the non-arbitrary organization of actions;
- and
- generating an output signal at least in dependence on said analyzing.
3 more elements
At least one processor, comprising:
- an input configured to receive a message containing information;
- a module configured to process the message with at least one trained artificial neural network-based processor, having at least one set of outputs which represent intelligent discriminations of the information at a respective order of intelligence complexity based on training, organized in a non-arbitrary organization of actions;
- a module configured to represent as a noise vector at least a portion of the information in the message which is incompletely represented in the non-arbitrary organization of actions as intelligent discriminations of the information;
- a module configured to analyze the noise vector distinctly from the at least one trained artificial neural network to extract characteristics of the at least one message distinct from the non-arbitrary organization of actions, having an output configured to provide a signal in dependence on said analysis of the noise vector.
1 more element
A method of processing information, comprising:
- receiving a message;
- providing a hierarchical stacked neural network having a plurality of layers, each layer of said hierarchal stacked neural network defining a respective transformation of received information implemented in an electronic processor, to produce an output pattern of a respective layer representing a non-arbitrary organization of actions at a respective stage of cognitive development, wherein respective layers of said hierarchal stacked neural network pass the respective non-arbitrary organization of actions at the respective stage of cognitive development to a successively higher layer having a successively higher stage of cognitive development, and respective layers of said hierarchal stacked neural network pass feedback to lower layers having respectively lower stages of cognitive development;
- processing the received message at a first layer of the hierarchical stacked neural network, to produce an output pattern of the first layer representing information corresponding to the message with respect to the respective stage of cognitive development of the first layer;
- and
- processing the output pattern of the first layer at a second layer of the hierarchical stacked neural network, to produce a neural network output pattern of the second layer representing a first portion of the information from the output pattern of the first layer corresponding to the non-arbitrary organization of actions at the respective stage of cognitive development of the second layer, and to extract a characteristic of the message from a noise vector output of the second layer representing at least a second portion of the information from the output pattern of the first layer not corresponding to the non-arbitrary organization of actions at the respective stage of cognitive development of the second layer and not corresponding to feedback to a respectively lower layer having a lower stage of cognitive development.
2 more elements
Dependencies are inferred from the claim text.