System and method for viewing, modifying, storing, and running artificial neural network components
AssigneeAudrey Kudritskiy
InventorAndrey KUDRITSKIY
Life of this patent
Estimating the term…
Abstract
A system and method for artificial neural network processing includes, for example, modifying, by a computer processor, a value of a charge of a node of an artificial neural depending on a number of elapsed steps since a prior predefined significant event. A system and method includes, for example, providing by a processor a real-time representation of an artificial neural network and of graphical effects of a running of the neural network. A system and method includes, for example, automatically modifying the behavior of network nodes based on simultaneous occurrences of events.
Claims25
A computer-implemented artificial neural network processing method, comprising:
- modifying, by a computer processor, a value of a charge of a node of an artificial neural network based on an input to the node, the node being configured to be activated to produce an output responsive to the value of the charge reaching a first threshold value;
- determining, by the processor, that a number of steps, that have elapsed since one of (a) a last activation of the node and (b) a last modification of the value of the charge based on input to the node, is greater than a second threshold value;
- and
- responsive to the determination, modifying, by the processor, the charge of the node.
A computer-implemented artificial neural network processing method, comprising:
- storing, by a computer processor, a plurality of node records, each including a respective identifier of a respective node, a respective charge value of the respective node, and a respective step identifier identifying a step at which the respective node one of had been activated and had its charge value modified in response to an input by an activated connection to which the respective node connects;
- storing, by the processor, a plurality of connection records, each including a respective identifier of a respective connection between a respective pair of nodes corresponding to a pair of the stored plurality of node records, a respective identifier of a first one of the respective pair of nodes from which the respective connection originates, a respective identifier of a second one of the respective pair of nodes at which the respective connection terminates, and a respective weight;
- and
- producing, by the processor, an output based on values identified in the node records and the connection records.
A computer-implemented artificial neural network visualization method, comprising:
- displaying, by a computer processor and in a first portion of a graphical user interface, a real-time graphical representation of the neural network;
- and
- displaying, by the processor and in a second portion of the graphical user interface, real-time graphical effects of a running of the neural network.
A computer-implemented method to at least one of learn and unlearn behavior in an artificial neural network, comprising:
- responsive to simultaneous activation of at least a first input node and a second input node connected to a third node, modifying, by a computer processor, a weight of a connection between the first input node and the third node.
A computer-implemented method to optimize an artificial neural network, comprising:
- modifying, by a computer processor, at least one of a) respective weight(s) of at least a connection of a neural network, and b) respective threshold(s) of at least a node of the neural network;
- responsive to a plurality of inputs to the neural network, determining whether at least one output of the neural network matches at least one desired output;
- upon determining that at least one output of the neural network matches at least one desired output, saving a configuration of the neural network and outputs of the neural network;
- repeating the modifying, the determining, and the saving steps until the at least one output of the neural network matches all desired outputs.
Dependencies are inferred from the claim text.