Granted Patent B2

US 12,260,321 · App. 17/385,316 · Granted Mar 25, 2025

Inventors: Fu-Cheng Tsai (Tainan, TW); Yi-Ching Kuo (Tainan, TW); Chih-Sheng Lin (Tainan, TW); Shyh-Shyuan Sheu (Zhubei, TW); Tay-Jyi Lin (Kaohsiung, TW); Shih-Chieh Chang (Hsinchu, TW)

Assignee: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

G06N3/063G06N3/04H03M1/1245G11C27/02H03M1/12

Abstract

Claims (37)

1. A data feature augmentation system for a low-precision neural network, comprising:

a first time difference unit, comprising:

a first sample-and-hold circuit, for receiving an input signal and obtaining a first signal according to the input signal, wherein the first signal is related to a first leakage rate of the first sample-and-hold circuit and the first signal is the signal generated by delaying the input signal by one time unit; and

a subtractor, for performing subtraction on the input signal and the first signal to obtain a time difference signal;

wherein the input signal and the time difference signal are inputted to the low-precision neural network.

2. The data feature augmentation system according to claim 1 , wherein the input signal, the first signal and the time difference signal are analog signals, and the data feature augmentation system further comprises:

a first analog-to-digital converter; and

a second analog-to-digital converter;

wherein the first analog-to-digital converter and the second analog-to-digital converter are low-resolution analog-to-digital converters, and before the input signal and the time difference signal are inputted to the low-precision neural network, the input signal and the time difference signal are respectively inputted to the first analog-to-digital converter and the second analog-to-digital converter.

3. The data feature augmentation system according to claim 1 , wherein the low-precision neural network is a binarized neural network or a short-wordlength neural network.

4. The data feature augmentation system according to claim 1 , wherein the first time difference unit further comprises:

a second sample-and-hold circuit, for receiving the input signal and obtaining a second signal according to the input signal, wherein the second signal is related to a second leakage rate of the second sample-and-hold circuit and the second signal is the signal generated by delaying the input signal by two time units;

wherein the subtractor is further used for subtracting the first signal and the second signal from the input signal to obtain the time difference signal.

5. The data feature augmentation system according to claim 4 , wherein the first sample-and-hold circuit and the second sample-and-hold circuit are connected in series.

6. The data feature augmentation system according to claim 4 , wherein the first sample-and-hold circuit and the second sample-and-hold circuit are connected in parallel.

7. The data feature augmentation system according to claim 4 , wherein the first leakage rate and the second leakage rate are adjustable.

8. The data feature augmentation system according to claim 1 , wherein the first time difference unit further comprises:

a second sample-and-hold circuit, for receiving the input signal and obtaining a second signal according to the input signal, wherein the second signal is related to a second leakage rate of the second sample-and-hold circuit and the second signal is the signal generated by delaying the input signal by two time units;

wherein the subtractor is further used for subtracting the second signal from the input signal to obtain the time difference signal.

9. The data feature augmentation system according to claim 8 , wherein the first sample-and-hold circuit and the second sample-and-hold circuit are connected in series.

10. The data feature augmentation system according to claim 8 , wherein the first sample-and-hold circuit and the second sample-and-hold circuit are connected in parallel.

11. The data feature augmentation system according to claim 1 , further comprising:

a second time difference unit, serially connected to the first time difference unit for obtaining a second time difference signal according to the time difference signal.

12. A data feature augmentation method for a low-precision neural network, comprising:

receiving an input signal and obtaining a first signal according to the input signal, wherein the first signal is related to a first leakage rate of a first sample-and-hold circuit and the first signal is the signal generated by delaying the input signal by one time unit; and

performing subtraction on the input signal and the first signal to obtain a time difference signal;

wherein the input signal and the time difference signal are inputted to the low-precision neural network.

13. The data feature augmentation method according to claim 12 , wherein the input signal, the first signal and the time difference signal are analog signals and the data feature augmentation method further comprises:

before the input signal and the time difference signal are inputted to the low-precision neural network, respectively inputting the input signal and the time difference signal to a first analog-to-digital converter and a second analog-to-digital converter, wherein the first analog-to-digital converter and the second analog-to-digital converter are low-resolution analog-to-digital converters.

14. The data feature augmentation method according to claim 12 , wherein the low-precision neural network is a binarized neural network or a short-wordlength neural network.

15. The data feature augmentation method according to claim 12 , further comprising:

receiving the input signal and obtaining a second signal according to the input signal, wherein the second signal is related to a second leakage rate of a second sample-and-hold circuit and the second signal is the signal generated by delaying the input signal by two time units; and

subtracting the first signal and the second signal from the input signal to obtain the time difference signal.

16. The data feature augmentation method according to claim 15 , wherein the first leakage rate and the second leakage rate are adjustable.

17. The data feature augmentation method according to claim 15 , further comprising:

receiving the input signal and obtaining a second signal according to the input signal, wherein the second signal is related to a second leakage rate of the second sample-and-hold circuit and the second signal is the signal generated by delaying the input signal by two time units; and

subtracting the second signal from the input signal to obtain the time difference signal.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (23)