Architecture for implementing an improved neural network

• US 10,235,621 B2
• Filed: 05/06/2014
• Issued: 03/19/2019
• Est. Priority Date: 05/07/2013
• Status: Active Grant

• Current Assignee: Iotelligent Technology Limited Incorporated
• Original Assignee: Iotelligent Technology Limited Incorporated
• Inventors: Zhu, Theodore
• Primary Examiner(s): Chace, Christian
• Assistant Examiner(s): Buss, Benjamin J
• Application Number:
  US14/271,137
• Publication Number:
  US 20140337263A1
• Time in Patent Office: 1,778 Days
• Field of Search: None
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: H04L 67/12 specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
• CPC Class Codes: H04W 4/38 for collecting sensor information
• CPC Class Codes: H04W 4/80 Services using short range communication, e.g. near-field communication [NFC], radio-frequency identification [RFID] or low energy communication
• CPC Class Codes: H04W 84/18 Self-organising networks, e.g. ad-hoc networks or sensor networks

Abstract

Disclosed is an improved approach to implement artificial neural networks. According to some approaches, an advanced neural network is implemented using an internet-of-things methodology, in which a large number of ordinary items having RFID technology are utilized as the vast infrastructure of a neural network.

Claims

1. An architecture for implementing a neural network, comprising:

a plurality of artificial neurons, the plurality of artificial neurons forming an arrangement of multiple devices that are networked together, wherein the plurality of the artificial neurons are networked together in a peer-to-peer network;

the plurality of artificial neurons include a control device, one or more sensor systems, and a plurality of memory devices;

the control device comprises a reader mechanism;

the one or more sensor systems perform sensing for at least one of speed, acceleration, temperature, moisture, sound, light capture, image capture, video capture, chemical detection, or GPS location;

the plurality of memory devices comprise a tag device having a memory component, wherein the tag device corresponds to a non-powered RFID tag that is a passive tag having a SRAM non-volatile memory;

the one or more sensor systems and the plurality of memory devices are within wireless communications range and wirelessly communicate with the control device, the arrangement of multiple devices that are networked together comprising billions of the artificial neurons forming trillions of connections; and

the plurality of artificial neurons are networked together into a neural network that includes a plurality of network pairings having a first artificial neuron that pairs with a second artificial neuron, wherein the plurality of artificial neurons are adaptively networked together in accordance with a plurality of rules comprising at least three rules, a first rule of the at least three rules establishing a star configuration having a plurality of leaf iotons connected to a central ioton, a second rule of the at least three rules establishing a tree configuration having a plurality of branches connected to a central ioton where a branch of the plurality of branches comprises a branch ioton and a leaf ioton, and a third rule of the at least three rules establishing a flat or web structure with each ioton connecting to any number of other iotons in the neural network directly or through one or more intermediate iotons, wherein the plurality of rules are applicable to form an arrangement of iotons such that a first set of iotons corresponding to the first rule forms the star configuration, a second set of iotons corresponding to the second rule forms the tree configuration, and a third set of iotons corresponding to the third rule forms the flat or web structures;

wherein individual iotons in the neural network exhibit different processing characteristics based upon positional location within the neural network, where a first ioton on an edge position of the neural network devotes a greater amount of processing power to computation in comparison to a second ioton in a central position of the neural network that functions as a relay or intermediary between multiple other iotons, where the second ioton devotes more processing power for communication as compared to the first ioton at the edge position of the neural network;

wherein the first artificial neuron communicates through an air interface in which a frequency range for the air interface is sub-divided into a plurality of frequency sub-ranges, such that the first artificial neuron is a part of a plurality of different neural networks that correspond to different ones of the plurality of frequency sub-ranges.

2. The architecture of claim 1, in which the peer-to-peer network comprises a mesh network.

3. The architecture of claim 1, in which the frequency range for the air interface corresponds to a 5 MHz range, and the 5 MHz range is sub-divided into the plurality of frequency sub-ranges that are 1.6 MHz sub-ranges.

4. The architecture of claim 1, in which the peer-to-peer network comprises an ad hoc pairing between two artificial neurons.

5. The architecture of claim 4, in which each artificial neuron is capable of pairing with another artificial neuron within communications range.

6. The architecture of claim 5, in which the pairing is implemented wirelessly using a wireless communications protocol.

7. The architecture of claim 6, in which the wireless communications protocol operates using a 900 MHz band using a 6lowpan protocol.

8. The architecture of claim 5, in which the pairing is implemented using a CDMA air interface.

9. The architecture of claim 8, in which the CDMA air interface is a TD-SCDMA interface with an operating frequency between 840 MHz and 845 MHz.

10. The architecture of claim 4, in which two artificial neurons can have different numbers of network pairings.

11. The architecture of claim 1, in which two artificial neurons can have different numbers or types of sensors.

12. The architecture of claim 1, in which the one or more sensor systems comprise a MEMS sensor.

13. The architecture of claim 1, in which the control device for the artificial neuron comprises a communications infrastructure that interfaces with the tag device using a wireless communications protocol.

14. The architecture of claim 13, in which the wireless communications protocol is implemented at a 900 MHz frequency.

15. The architecture of claim 13, in which the wireless communications protocol is implemented using an RFID protocol.

16. The architecture of claim 13, in which the control device for the first artificial neuron further comprises a communications infrastructure that interfaces with the one or more sensor systems using a wireless communications protocol.

17. The architecture of claim 16, in which the control device interfaces with the one or more sensor systems using a ZigBee protocol.

18. The architecture of claim 16, in which the one or more sensor systems are part of or integral with the tag device, such that the control device is able to interface with the one or more sensor systems using an RFID protocol.

19. The architecture of claim 13, in which the control device and a memory device are integrated into a single device with both controlling and memory functions.

20. The architecture of claim 19, in which the single device having both the control device and the memory device is implemented as a single IC chip having the reader mechanism and the one or more sensor systems.

21. The architecture of claim 1, in which the reader mechanism and the one or more sensor systems are implemented as separate components in the control device.

22. The architecture of claim 1, in which the control device and the tag devices are distributed as objects in an internet-of-things architecture.

23. The architecture of claim 1, in which the control device and the tag devices are distributed in consumer items.

24. The architecture of claim 1, in which the neural network implements decision-making tools, pattern-matching algorithms, learning systems, or control systems.

26. The architecture of claim 1, where the first artificial neuron is part of a first neural network corresponding to a first frequency sub-range and is part of a second neural network corresponding to a second frequency sub-range.

25. A method for implementing a neural network, comprising:

distributing a plurality of artificial neurons as distributed devices, wherein the plurality of artificial neurons comprise short range communications technology,

implementing the plurality of artificial neurons to include a control device, one or more sensor systems, and a plurality of memory devices;

providing the control device to include a reader mechanism;

providing the one or more sensor systems to perform sensing for at least one of speed, acceleration, temperature, moisture, sound, light capture, image capture, video capture, chemical detection, or GPS location;

providing the plurality of memory devices to comprise a tag device having a memory component, wherein the tag device corresponds to a non-powered RFID tag that is a passive tag having a SRAM non-volatile memory;

networking the plurality of artificial neurons together into an arrangement comprising billions of the artificial neurons forming trillions of connections, wherein the plurality of artificial neurons are networked together as part of an adaptable peer-to-peer network;

adaptively networking the plurality of artificial neurons together in accordance with a plurality of rules comprising at least three rules, a first rule of the at least three rules establishing a star configuration having a plurality of leaf iotons connected to a central ioton, a second rule of the at least three rules establishing a tree configuration having a plurality of branches connected to a central ioton where a branch of the plurality of branches comprises a branch ioton and a leaf ioton, and a third rule of the at least three rules establishing a flat or web structure with each ioton connecting to any number of other iotons in the neural network directly or through one or more intermediate iotons, wherein the plurality of rules are applicable to form an arrangement of iotons such that a first set of iotons corresponding to the first rule forms the star configuration, a second set of iotons corresponding to the second rule forms the tree configuration, and a third set of iotons corresponding to the third rule forms the flat or web structure;

implementing different processing characteristics for individual iotons in the neural network based upon positional location within the neural network, where a first ioton on an edge position of the neural network devotes a greater amount of processing power to computation in comparison to a second ioton in a central position of the neural network that functions as a relay or intermediary between multiple other iotons, where the second ioton devotes more processing power for communication as compared to the first ioton at the edge position of the neural network; and

configuring an artificial neuron to communicate through an air interface in which a frequency range for the air interface is sub-divided into a plurality of frequency sub-ranges, such that the artificial neuron is a part of a plurality of different neural networks that correspond to different ones of the plurality of frequency sub-ranges.

27. The method of claim 25, in which the control device comprises an RFID reader mechanism, and the plurality of memory devices comprise a RFID tag having memory component that is within communications range of the control device.

28. The method of claim 27, in which the control device discovers the RFID tag within the communications range and uses the RFID tag as its associated memory device.

29. The method of claim 25, in which the plurality of artificial neurons form the peer-to-peer network to implement processing functions.

30. The method of claim 29, in which the peer-to-peer network comprises a mesh network.

31. The method of claim 25, in which two artificial neurons can have different numbers of network pairings.

32. The method of claim 25, in which the plurality of artificial neurons form pairings based upon input received through a sensor.

33. The method of claim 32, in which the plurality of artificial neurons can have different numbers or types of sensors.

34. The method of claim 33, in which at least one sensor comprises a MEMS sensor.

35. The method of claim 25, in which the plurality of artificial neurons are distributed as objects in an internet-of-things architecture.

36. The method of claim 25, in which the control device and RFID tags of the plurality of artificial neurons are distributed in consumer items.

37. The method of claim 25, in which the neural network implements decision-making tools, pattern-matching algorithms, learning systems, or control systems.

38. The method of claim 25, where the artificial neuron is part of a first neural network corresponding to a first frequency sub-range and is part of a second neural network corresponding to a second frequency sub-range.