Title:
INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2024/048265
Kind Code:
A1
Abstract:
The present disclosure relates to an information processing device, an information processing method, and a program that make it possible to detect forgery data effectively using a more versatile technique. The present invention involves calculating contribution degrees which indicate the degrees of contribution, to prediction labels outputted from a trained model, made by respective feature amounts in a training data set that is formed as a mixture of normal samples which include only normal data and forgery samples which at least partially include forgery data. Then, by using an unsupervised learning which uses the contribution degrees as inputs and by comparing variation of the feature amounts in respective clusters in a clustering result obtained by executing clustering that divides samples of the training data set into a plurality of clusters, a cluster in which the forgery samples included in the training data set are classified is identified. The present technology can be applied to a machine learning system which generates a fraud detection model, for example.
Inventors:
FUNABIKI YUKI (JP)
CARTELLA FRANCESCO (BE)
ANUNCIACAO ORLANDO (BE)
CVETKO VAH KARIN (BE)
CARTELLA FRANCESCO (BE)
ANUNCIACAO ORLANDO (BE)
CVETKO VAH KARIN (BE)
Application Number:
PCT/JP2023/029488
Publication Date:
March 07, 2024
Filing Date:
August 15, 2023
Export Citation:
Assignee:
SONY GROUP CORP (JP)
International Classes:
G06N20/00
Domestic Patent References:
| WO2021111540A1 | 2021-06-10 |
Foreign References:
| US20210081831A1 | 2021-03-18 | |||
| US20210081708A1 | 2021-03-18 |
Other References:
SCHULTH LUKAS, BERGHOFF CHRISTIAN, NEU MATTHIAS: "Detecting Backdoor Poisoning Attacks on Deep Neural Networks by Heatmap Clustering", ARXIV (CORNELL UNIVERSITY), CORNELL UNIVERSITY LIBRARY, ARXIV.ORG, ITHACA, 27 April 2022 (2022-04-27), Ithaca, XP093145115, [retrieved on 20240325], DOI: 10.48550/arxiv.2204.12848
Attorney, Agent or Firm:
NISHIKAWA Takashi et al. (JP)
Download PDF: