Intelligent control with hierarchical stacked neural networks
First Claim
1. At least one processor configured to automatically carry out a sensory-motor task, comprising:
- an input configured to receive information;
a hierarchical stacked neural network comprising at least three hierarchically interconnected neural network layers, each neural network layer above a lowest hierarchical level comprising at least a plurality of input neurons receiving a plurality of respective layer inputs from a preceding hierarchical level neural network layer, which influence a plurality of output neurons of the respective neural network layer to produce a plurality of respective layer outputs, each respective neural network layer having a plurality of parameters representing training of the respective neural network layer, wherein the respective layer outputs comprise a non-arbitrary organization of actions, each respective neural network layer being configured to interpret the information according to successive stages of cognitive development, and respectively higher neural network layers being configured to provide feedback to respectively lower neural network layers to adjust the plurality of parameters;
an interface configured to communicate a query for search of at least one external database defined by an output of at least one layer of the hierarchical stacked neural network, and for returning data corresponding to the search to at least one layer of the hierarchical stacked neural network higher than the at least one layer of the hierarchical stacked neural network which defined the query, for processing therein; and
a motor network configured to control the sensory-motor task based on an output from the hierarchical stacked neural network, being dependent on the returned data.
0 Assignments
0 Petitions
Accused Products
Abstract
A method of processing information is provided. The method involves receiving a message; processing the message with a trained artificial neural network based processor, having at least one set of outputs which represent information in a non-arbitrary organization of actions based on an architecture of the artificial neural network based processor and the training; representing as a noise vector at least one data pattern in the message which is incompletely represented in the non-arbitrary organization of actions; analyzing the noise vector distinctly from the trained artificial neural network; searching at least one database; and generating an output in dependence on said analyzing and said searching.
-
Citations
21 Claims
-
1. At least one processor configured to automatically carry out a sensory-motor task, comprising:
-
an input configured to receive information; a hierarchical stacked neural network comprising at least three hierarchically interconnected neural network layers, each neural network layer above a lowest hierarchical level comprising at least a plurality of input neurons receiving a plurality of respective layer inputs from a preceding hierarchical level neural network layer, which influence a plurality of output neurons of the respective neural network layer to produce a plurality of respective layer outputs, each respective neural network layer having a plurality of parameters representing training of the respective neural network layer, wherein the respective layer outputs comprise a non-arbitrary organization of actions, each respective neural network layer being configured to interpret the information according to successive stages of cognitive development, and respectively higher neural network layers being configured to provide feedback to respectively lower neural network layers to adjust the plurality of parameters; an interface configured to communicate a query for search of at least one external database defined by an output of at least one layer of the hierarchical stacked neural network, and for returning data corresponding to the search to at least one layer of the hierarchical stacked neural network higher than the at least one layer of the hierarchical stacked neural network which defined the query, for processing therein; and a motor network configured to control the sensory-motor task based on an output from the hierarchical stacked neural network, being dependent on the returned data. - View Dependent Claims (2, 3, 4, 5, 6, 7)
-
-
8. A method of automatically performing a sensory-motor task, comprising:
-
receiving an input comprising sensory information from an environment; interpreting the sensory information in a hierarchical stacked neural network comprising at least three layers, each layer comprising at least a plurality of input neurons which influence a plurality of output neurons, and being associated with a plurality of weights, each layer producing a non-arbitrary organization of actions representing a level of cognitive development respectively higher than a level of cognitive development represented in inputs to the respective layer, and respectively higher layers providing feedback to respectively lower levels; searching at least one external database comprising information relevant to at least one of the environment and the sensory-motor task, based on an output of at least one layer of the hierarchical stacked neural network; receiving a response from the searching of the at least one external database by the hierarchical stacked neural network as an input of at least one layer at a respectively higher layer than the output of the at least one layer that produced the output; processing the received response by at least the hierarchical stacked neural network to define a sensory-motor effect; and providing an output comprising a representation of effector signals for controlling the sensory-motor task. - View Dependent Claims (9, 10, 11, 12, 13, 14, 15)
-
-
16. A processor configured to carry out a sensory-motor task, comprising:
-
an input configured to receive environmental information; a hierarchical stacked neural network comprising at least three layers operating at different levels of cognitive development; each layer comprising at least one programmable inter-neuron weight and each layer below a highest layer receiving feedback from a respective higher layer, and being configured to interpret abstract information represented in the environmental information and to produce a non-arbitrary organization of actions dependent on the respective level of cognitive development; a module configured to conduct an external database search controlled by at least the hierarchical stacked neural network representing at least a non-arbitrary organization of actions of a query layer of the hierarchical stacked neural network based on a goal of the sensory-motor task and the environmental information, a response to the external database search being provided as an input to the hierarchical stacked neural network and being processed by at least one layer of the hierarchical stacked neural network above a level of the query layer; and a module configured to generate control information relating to a manner of carrying out the sensory-motor task based at least on an output of the hierarchical stacked neural network dependent on a result of the module configured to conduct an external database search. - View Dependent Claims (17, 18, 19, 20, 21)
-
Specification