US 12,266,152 B2
Method and system of designing memristor-based naive Bayes classifier and classifier
Huajun Sun, Hubei (CN); Zuopai Zhou, Hubei (CN); Li Li, Hubei (CN); and Xiangshui Miao, Hubei (CN)
Assigned to HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY, Hubei (CN)
Appl. No. 17/775,591
Filed by HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY, Hubei (CN)
PCT Filed May 7, 2021, PCT No. PCT/CN2021/092106
§ 371(c)(1), (2) Date May 10, 2022,
PCT Pub. No. WO2022/217673, PCT Pub. Date Oct. 20, 2022.
Claims priority of application No. 202110403525.X (CN), filed on Apr. 14, 2021.
Prior Publication US 2023/0334824 A1, Oct. 19, 2023
Int. Cl. G06V 10/84 (2022.01); G06N 3/06 (2006.01); G06V 10/764 (2022.01); G06V 10/94 (2022.01)
CPC G06V 10/764 (2022.01) [G06V 10/955 (2022.01)] 13 Claims
OG exemplary drawing
 
1. A method of designing a memristor-based naive Bayes classifier, comprising:
S1: constructing a naive Bayes classifier comprising a memristor array of M rows by 2N columns, where M is the number of types outputted by the naive Bayes classifier, N is the number of pixels in a picture, and the pixel value of each pixel is 0 or 1;
S2: selecting a jth training sample corresponding to a jth type from training samples and calculating the number hj,2i-1 of the pixel value of 0 and the number hj,2i of the pixel value of 1 in an ith pixel in the jth training sample, where j=1, 2, . . . , and M, and i=1, 2, . . . , and N; and
S3: applying hj,2i-1 pulses to a memristor Rj,2i-1 in a jth row and a 2i−1th column to modulate the conductance of the memristor Rj,2i-1 and applying hj,2i pulses to a memristor Rj,2i in the jth row and a 2ith column to modulate the conductance of the memristor Rj,2i,
wherein the memristor-based naive Bayes classifier is obtained through S1 to S3,
wherein during modulation, the conductance of each memristor is:
G=a×ln(N1)+b, and
a classification statistical calculation probability corresponding to the memristor and the conductance of the memristor satisfy:
ln P∝−G
wherein G is the conductance of the memristor, N1 is the number of pulses applied to the memristor, P is a conditional probability value corresponding to the memristor, ln P is the classification statistical calculation probability, a is a first fitting parameter, and b is a second fitting parameter.