초록 ▼

This application discloses a neural network that also functions as a connection oriented packet data network using an MPLS-type label switching technology. The neural network uses its intelligence to build and manage label switched paths (LSPs) to transport user packets and solve complex mathematica

This application discloses a neural network that also functions as a connection oriented packet data network using an MPLS-type label switching technology. The neural network uses its intelligence to build and manage label switched paths (LSPs) to transport user packets and solve complex mathematical problems. However, the methods taught here can be applied to other data networks including ad-hoc, mobile, and traditional packet networks, cell or frame-switched networks, time-slot networks and the like.

대표청구항 ▼

1. A method of adding neural network functionality to an array of four or more interconnected nodes of a packet data network where each node of the array is capable of communicating with at least two other nodes, and where a node corresponds to one or more neurons of the neural network, comprising:

1. A method of adding neural network functionality to an array of four or more interconnected nodes of a packet data network where each node of the array is capable of communicating with at least two other nodes, and where a node corresponds to one or more neurons of the neural network, comprising: receiving a path request packet at a first node of the array from a node in communication with the first node, where the request packet is used in the path building process;using a portion of the path request packet to access a time delay value, where the time delay value serves as a weight function for the neural network and is used in the path selection process;initiating a packet waiting time delay based on the retrieved value; andtransmitting at least a portion of the request packet to at least a second node to expose a possible path for use by the corresponding neural network, after the packet waiting time has expired. 2. The method of claim 1 where the nodes are on a semiconductor substrate. 3. The method of claim 1 where the nodes are on a printed circuit board (PCB). 4. The method of claim 1 where the packet data network is connection oriented. 5. The method of claim 4 where the path is a label switched path (LSP). 6. The method of claim 1 where the request packet is a hunting packet. 7. The method of claim 1 where the packet data network is a data neural network. 8. A method of using the neural network functionality of claim 1 to solve a problem containing multiple related objects comprising: organizing the objects of the problem into an array that is smaller than the array of claim 1;mapping the objects onto selected nodes of claim 1;converting the relationship between the objects for the first node and second node of claim 1 into a packet waiting time delay value; andstoring the value into a memory of claim 1. 9. A neural network containing an array of four or more interconnected nodes of a packet data network where each node of the array corresponds to one or more neurons of the neural network, and is capable of communicating with at least two other nodes; the packet data network containing a first node with software or circuitry capable of receiving a path request packet from a node in communication with the first node, where the request packet is to be used in the path building process; the first node containing software or circuitry capable of initiating a packet waiting time delay which serves as a weight in the neural network, and is based on information included with the request packet, combined with information stored in a memory, and where the weight is also capable of influencing the paths selected by packets traversing the network;the first node further containing software or circuitry for transmitting at least a portion of the request packet to at least a second node to expose a possible path for use by the corresponding neural network, after the packet waiting time has expired. 10. A method of synthesizing a neural network from a packet data network containing a plurality of interconnected nodes where each node corresponds to one or more neurons of the neural network, comprising: receiving at a first node a path request from another node of the network, where the path request is used in the path selection process; initiating a time delay in response to the reception, where the time delay value is derived from data in a memory, and where the delay is used as a weight function for the corresponding neural network, and is also used to influence the routing of packets through the network; andtransmitting at least a portion of the packet to at least a second node of the network to expose a possible path for use by the corresponding neural network, after the expiration of the time delay.