NEURAL-NETWORK-BASED CLASSIFICATION DEVICE AND CLASSIFICATION METHOD

• US 20200134393A1
• Filed: 12/17/2018
• Published: 04/30/2020
• Est. Priority Date: 10/29/2018
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Lai, Ching-Hao, Huang, Mao-Yu
• Granted Patent: US 11,704,534 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/2113 by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/245 relating to the decision surface
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology

Abstract

Provided is a neural-network-based classification method, including: generating, by a neural network, one or more score vectors corresponding to one or more samples respectively; determining a first subset of the one or more samples according to the one or more score vectors and a first decision threshold, wherein the first subset is associated with a first class; and selecting samples to be re-examined from the one or more samples according to the first subset.

Claims

1. A neural-network-based classification device, comprising:

a storage medium, storing a plurality of modules; and

a processor, coupled to the storage medium, the processor accessing and executing the plurality of modules, the plurality of modules comprising;

a neural network, generating one or more score vectors corresponding to one or more samples respectively;

a classifier, determining a first subset of the one or more samples according to the one or more score vectors and a first decision threshold, wherein the first subset is associated with a first class; and

a computation module, selecting samples to be re-examined from the one or more samples according to the first subset.

2. The neural-network-based classification device according to claim 1, wherein

the classifier further determines a second subset of the one or more samples according to the one or more score vectors and a second decision threshold, wherein the second subset is associated with a second class;

and

the computation module further selects the samples to be re-examined from the one or more samples according to the first subset and the second subset.

3. The neural-network-based classification device according to claim 2, wherein

the first decision threshold is configured to make samples in the first subset satisfy a first requirement;

and

the second decision threshold is configured to make samples in the second subset satisfy a second requirement.

4. The neural-network-based classification device according to claim 2, wherein the samples to be re-examined are included in a complement of a union of the first subset and the second subset in the one or more samples.

5. The neural-network-based classification device according to claim 1, wherein the first decision threshold is associated with at least one of an accuracy and a precision rate.

6. The neural-network-based classification device according to claim 1, wherein the classifier is configured to execute:

determining a third subset of the one or more samples according to the one or more score vectors and a third decision threshold, wherein the third subset is associated with the first class; and

determining the first subset from the third subset according to the one or more score vectors and the first decision threshold.

7. The neural-network-based classification device according to claim 6, wherein the third decision threshold is configured to make samples in the third subset satisfy a third requirement.

8. The neural-network-based classification device according to claim 6, wherein the samples to be re-examined are included in a complement of the first subset in the third subset.

9. The neural-network-based classification device according to claim 6, wherein the third decision threshold is associated with at least one of a recall rate and a miss rate.

10. The neural-network-based classification device according to claim 1, wherein

the classifier further determines a third subset of the one or more samples according to the one or more score vectors and a third decision threshold, wherein the third subset is associated with the first class;

and

the computation module further selects the samples to be re-examined from the one or more samples according to the first subset and the third subset.

11. The neural-network-based classification device according to claim 10, wherein the third decision threshold is configured to make samples in the third subset satisfy a third requirement.

12. The neural-network-based classification device according to claim 10, wherein the samples to be re-examined are included in a complement of the first subset in the third subset.

13. The neural-network-based classification device according to claim 10, wherein the third decision threshold is associated with at least one of a recall rate and a miss rate.

14. A neural-network-based classification method, comprising:

generating, by a neural network, one or more score vectors corresponding to one or more samples respectively;

determining a first subset of the one or more samples according to the one or more score vectors and a first decision threshold, wherein the first subset is associated with a first class; and

selecting samples to be re-examined from the one or more samples according to the first subset.

15. The neural-network-based classification method according to claim 14, further comprising:

determining a second subset of the one or more samples according to the one or more score vectors and a second decision threshold, wherein the second subset is associated with a second class; and

selecting the samples to be re-examined from the one or more samples according to the first subset and the second subset.

16. The neural-network-based classification method according to claim 15, wherein

the first decision threshold is configured to make samples in the first subset satisfy a first requirement;

and

the second decision threshold is configured to make samples in the second subset satisfy a second requirement.

17. The neural-network-based classification method according to claim 15, wherein the samples to be re-examined are included in a complement of a union of the first subset and the second subset in the one or more samples.

18. The neural-network-based classification method according to claim 14, wherein the first decision threshold is associated with at least one of an accuracy and a precision rate.

19. The neural-network-based classification method according to claim 14, wherein the step of determining the first subset of the one or more samples according to the one or more score vectors and the first decision threshold comprises:

determining a third subset of the one or more samples according to the one or more score vectors and a third decision threshold, wherein the third subset is associated with the first class; and

determining the first subset from the third subset according to the one or more score vectors and the first decision threshold.

20. The neural-network-based classification method according to claim 19, wherein the third decision threshold is configured to make samples in the third subset satisfy a third requirement.

21. The neural-network-based classification method according to claim 19, wherein the samples to be re-examined are included in a complement of the first subset in the third subset.

22. The neural-network-based classification method according to claim 19, wherein the third decision threshold is associated with at least one of a recall rate and a miss rate.

23. The neural-network-based classification method according to claim 14, further comprising:

determining a third subset of the one or more samples according to the one or more score vectors and a third decision threshold, wherein the third subset is associated with the first class; and

selecting the samples to be re-examined from the one or more samples according to the first subset and the third subset.

24. The neural-network-based classification method according to claim 23, wherein the third decision threshold is configured to make samples in the third subset satisfy a third requirement.

25. The neural-network-based classification method according to claim 23, wherein the samples to be re-examined are included in a complement of the first subset in the third subset.

26. The neural-network-based classification method according to claim 23, wherein the third decision threshold is associated with at least one of a recall rate and a miss rate.