SYSTEM AND METHOD FOR DETERMINING AN ARTIFICIAL INTELLIGENCE MODEL IN A DECENTRALIZED NETWORK

• US 20190311246A1
• Filed: 04/10/2018
• Published: 10/10/2019
• Est. Priority Date: 04/10/2018
• Status: Active Grant

• Current Assignee: Gyrfalcon Technology Inc.
• Original Assignee: Gyrfalcon Technology Inc.
• Inventors: Yang, Lin, Young, Charles Jin, Dong, Jason Zeng, Dong, Patrick Zeng, Sun, Baohua, Zhang, Yequn
• Granted Patent: US 10,943,168 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A system may include a decentralized communication network and multiple processing devices on the network. Each processing device may have an artificial intelligence (AI) chip, the device may be configured to generate an AI model, determine the performance value of the AI model on the AI chip, receive a chain from the network where the chain contains a performance measure. If the performance value of the AI model is better than the performance measure, then the processing device may broadcast the AI model to the network for verification. If the AI model is verified by the network, the device may update the chain with the performance value so that the chain can be shared by the multiple processing devices on the network. Any processing device on the network may also verify an AI model broadcasted by any other device. Methods for generating the AI model are also provided.

Claims

1. A device comprising:

an artificial intelligence (AI) chip; and

a processing device containing programming instructions that, when executed, will cause the processing device to;

(i) access a dataset;

(ii) generate a first artificial intelligence (AI) model;

(iii) load the first AI model into the AI chip to generate a first performance value based on the dataset;

(iv) receive a chain from a communication network, the chain comprising a performance measure;

(v) determine whether the first performance value has exceeded the performance measure; and

(vi) if the first performance value has exceeded the performance measure;

create a block comprising the first AI model and the first performance value;

transmit the block and a verification request to the communication network to verify the block; and

if verification of the block succeeds, update the chain based on the first performance value so that the chain is available to a plurality of additional processing devices on the communication network.

2. The device of claim 1, wherein the first AI model is a convolutional neural network comprising a plurality of weights for loading into the AI chip.

3. The device of claim 1 further comprise additional programming instructions that will cause the processing device to:

receive from the communication network the chain comprising

receive from the communication network an additional verification request and a second block comprising a second AI model;

respond to the additional verification request from the communication network by;

loading the second AI model into the AI chip to generate a second performance value based on the dataset;

determining whether the second performance value has exceeded the performance measure;

if the second performance value has exceeded the performance measure, responding to the additional verification request via the communication network with a verification success.

4. The device of claim 1, wherein programming instructions for verifying the block comprise programming instructions that will:

(vii) receive responses from one or more of the additional plurality of processing devices responding to the verification request; and

(viii) based on the received responses, determine that the verification of the block is successful if at least a percentage of the total number of processing devices receiving the verification request has responded with a verification success.

5. The device of claim 4, wherein programming instructions for verifying the block also contain programming instructions that will:

repeat steps (vii)-(viii) within a time period; and

determine that the verification of the block fails if not more than the percentage of the total number of processing devices receiving the verification request has responded with a verification success within the time period.

6. The device of claim 3, wherein programming instructions for receiving the chain that comprises the performance measure arc configured to receive the performance measure at a time interval.

7. The device of claim 1 further comprising additional programming instructions configured to:

if verification of the block is successful, receive an award if no chain that comprises a performance value that is better than the first performance value is received on the communication network within a time period.

8. The device of claim 1, wherein programming instructions for generating the first AI model comprise additional programming instructions configured to convert a floating model to a fixed point model.

9. The device of claim 1, wherein programming instructions for generating the first AI model comprise programming instructions configured to:

(i) initialize the first AI model;

(ii) determine a string;

(iii) use a hash encoder to generate a hash code based on the string;

(iv) fill in the first AI model with the hash code;

(v) determine whether the first AI model has been filled up;

(vi) if the first AI model has not been filled up;

update the string with the hash code;

use the hash encoder to generate the hash code based on the updated string; and

repeat steps (iv)-(vi) until the first AI model has been filled up.

10. The device of claim 9, wherein the string comprises a unique random string.

11. A method comprising, by a processing device:

(i) accessing a dataset;

(ii) generating a first artificial intelligence (AI) model;

(iii) loading the first AI model into an AI chip to generate a first performance value based on the dataset;

(iv) receiving a chain from a communication network, the chain comprising a performance measure;

(v) determining whether the first performance value has exceeded the performance measure; and

(vi) if the first performance value has exceeded the performance measure;

creating a block comprising the first AI model and the first performance value;

transmitting the block and a verification request to the communication network to verify the block; and

if verification of the block succeeds, updating the chain based on the first performance value so that the chain is available to a plurality of additional processing devices on the communication network.

12. The method of claim 11, wherein the first AI model is a convolutional neural network comprising a plurality of weights for loading into the AI chip.

13. The method of claim 11 further comprising:

receiving from the communication network the chain comprising the performance value;

receiving from the communication network an additional verification request and a block comprising a second AI model;

responding to the additional verification request from the communication network by;

loading the second AI model into the AI chip to generate a second performance value based on the dataset;

determining whether the second performance value has exceeded the performance measure;

if the second performance value has exceeded the performance measure, responding to the additional verification request via the communication network with a verification success.

14. The method of claim 11, wherein verifying the block comprises:

(vii) receiving responses from one or more of the additional plurality of processing devices responding to the verification request; and

(viii) based on the received responses, determining that the verification of the block is successful if at least a percentage of the total number of processing devices receiving the verification request has responded with a verification success.

15. The method of claim 14, wherein verifying the block further comprising:

repeating steps (vii)-(viii) within a time period; and

determining that the verification of the block fails if not more than the percentage of the total number of processing devices receiving the verification request has responded with a verification success within the time period.

16. The method of claim 13, wherein receiving the chain comprising the performance measure occurs at a time interval.

17. The method of claim 11, further comprising:

if verification of the block is successful, receiving an award if no chain that comprises a performance value that is better than the first performance value is received on the communication network within a time period.

18. The method of claim 11, wherein generating the first AI model comprises converting a floating model to a fixed point model.

19. The method of claim 11, wherein generating the first AI model comprises:

(i) initializing the first AI model;

(ii) determining a string;

(iii) using a hash encoder to generate a hash code based on the string;

(iv) filling in the first AI model with the hash code;

(v) determining whether the first AI model has been filled up;

(vi) if the first AI model has not been filled up;

updating the string with the hash code;

using the hash encoder to generate the hash code based on the updated string; and

repeating steps (iv)-(vi) until the first AI model has been filled up.

20. The method of claim 19, wherein the string comprises a unique random string.