US 12,450,521 B2
Learning device, learning method, and storage medium for learning device
Shuhei Yoshida, Tokyo (JP); and Makoto Terao, Tokyo (JP)
Assigned to NEC CORPORATION, Tokyo (JP)
Appl. No. 17/801,554
Filed by NEC Corporation, Tokyo (JP)
PCT Filed Mar. 13, 2020, PCT No. PCT/JP2020/011157
§ 371(c)(1), (2) Date Aug. 23, 2022,
PCT Pub. No. WO2021/181668, PCT Pub. Date Sep. 16, 2021.
Prior Publication US 2023/0177389 A1, Jun. 8, 2023
Int. Cl. G06N 20/00 (2019.01)
CPC G06N 20/00 (2019.01) 6 Claims
OG exemplary drawing
 
1. A learning device comprising:
at least one memory configured to store instructions; and
at least one processor configured to execute the instructions to:
calculate a recognition loss using: a recognition result with respect to recognition object data in a learning data set that is a set of a pair of the recognition object data and a weak label; a mixing matrix calculated based on the learning data set; and the weak label attached to the recognition object data, wherein the learning data set comprises a weak label probability distribution,
the weak label probability distribution is a probability distribution followed by the weak label conditioned by a true correct answer class to which the recognition object data belongs, and has a left inverse matrix when the weak label is regarded as a row index and the true correct answer class is regarded as a column index,
calculating the recognition loss by the at least one processor comprises:
calculating a difference between the mixing matrix and the recognition result; and
calculating the recognition loss by calculating a sum of a square of the difference, and
the at least one processor is configured to execute the instructions to further update a parameter of a recognition model based on the recognition loss.