LEARNING DEVICE, SEARCH DEVICE, LEARNING METHOD, SEARCH METHOD, AND PROGRAM

WIPO Patent Application WO/2021/215042

A1

A learning device that generates a feature vector using a model parameter of a neural network, the learning device being characterized by comprising: a conversion unit that performs conversion of a plurality of first feature vectors, a plurality of second feature vectors, and a plurality of third feature vectors respectively into a plurality of first learning feature vectors, a plurality of second learning feature vectors, and a plurality of third learning feature vectors, the plurality of first feature vectors being of a search query, the plurality of second feature vectors being of a first document related to the search query, the plurality of third feature vectors being of a second document unrelated to the search query, the conversion being performed by adjusting, for each of the plurality of first feature vectors, the plurality of second feature vectors, and the plurality of third feature vectors, the number of elements that have zero in each dimension on the basis of a predetermined condition; and an update unit that updates the model parameter using the plurality of first learning feature vectors, the plurality of second learning feature vectors, and the plurality of third learning feature vectors.

HASEGAWA TAKU (JP)
NISHIDA KYOSUKE (JP)
KAKU SOICHIRO (JP)
YOSHIDA SEN (JP)
TOMITA JUNJI (JP)

PCT/JP2020/045919

October 28, 2021

December 09, 2020

Click for automatic bibliography generation  

NIPPON TELEGRAPH & TELEPHONE (JP)

G06F16/31; G06F16/38

JP2019101993A	2019-06-24
JP2018032232A	2018-03-01
JP2019053645A	2019-04-04

ITOH, Tadashige et al. (JP)

View/Download PDF      PDF Help

Previous Patent: POWER AMPLIFIER MODULE AND COMMUNICATION DEVICE

Next Patent: WIRE ROPE MONITORING DEVICE AND CRANE SYSTEM