CLASSIFICATION DEVICE AND METHOD WITH REDUCED BIAS

Title:

CLASSIFICATION DEVICE AND METHOD WITH REDUCED BIAS

Document Type and Number:

WIPO Patent Application WO/2020/149450

Kind Code:

A1

Abstract:

The present invention provides a classification device and a classification method, the classification device comprising: a data compression unit for extracting feature values from input data including attribute values for a plurality of attributes according to a pre-learned pattern recognition method and transferring the feature values to a latent space of a pre-specified dimension; and a classification unit for classifying the feature values transferred to the latent space according to a pre-learned pattern recognition method separately from the data compression unit and reconstructing data classified from the classified feature values, wherein the data compression unit is learned by inversely propagating a bias error measured by analyzing, during learning, a transfer error, which is an error between the data reconstructed from the feature values transferred to the latent space and the input data, and a statistical characteristic of a latent variable corresponding to a pre-specified attribute from the feature values transferred to the latent space.

More Like This:

WO/2022/020006	COMPRESSING TOKENS BASED ON POSITIONS FOR TRANSFORMER MODELS
WO/2020/040829	HYBRID PHYSICS-BASED AND MACHINE LEARNING MODELS FOR RESERVOIR SIMULATIONS
JP2022510712	Neural network training method and image matching method, as well as equipment

Inventors:

CHO SUNG-BAE (KR)
KIM JIN-YOUNG (KR)
CHOO SHANE (KR)
CHO YU SEONG (KR)

Application Number:

PCT/KR2019/002698

Publication Date:

July 23, 2020

Filing Date:

March 08, 2019

Export Citation:

Click for automatic bibliography generation Help

Assignee:

UNIV YONSEI IACF (KR)

International Classes:

G06N3/08; G06F16/35

Foreign References:

US20170330058A1	2017-11-16
JPS5139874B2	1976-10-30
US20180314917A1	2018-11-01

Other References:

KIM, JIN-YOUNG ET AL.: "A Latent Space Matching Model Based on Deep Autoencoder for Reducing Algorithmic Bias of Gender", KOREA SOFTWARE CONGRESS 2018, 21 December 2018 (2018-12-21)
ZHAO, JIEYU ET AL.: "Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints", ARXIV: 1707.09457V1, 29 July 2017 (2017-07-29), XP080959841, Retrieved from the Internet [retrieved on 20190920]

Attorney, Agent or Firm:

MIN, Young Joon (KR)

Download PDF:

View/Download PDF PDF Help

Previous Patent: ABSORBENT BONE HEMOSTATIC MATERIAL COMPOSITION, AND BONE HEMOSTATIC MATERIAL MANUFACTURING METHOD US...

Next Patent: THREE-DIMENSIONAL POROUS STRUCTURE AND FABRICATION METHOD THEREFOR