Deep learning apparatus and method able to be used for anomaly detection

In the field of anomaly detection, unsupervised learning is used rather than widely used supervised learning because of the difficulty of data labeling and the imbalance of datasets. The unsupervised learning does not require labeling, but the performance of the unsupervised learning is inferior to...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: PARK HEA SOOK, CHUNG BYUNG CHANG, JUNG BOO GEUM, YIM JIN HYUK, YOO YOON SIK
Format: Patent
Sprache:eng ; kor
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In the field of anomaly detection, unsupervised learning is used rather than widely used supervised learning because of the difficulty of data labeling and the imbalance of datasets. The unsupervised learning does not require labeling, but the performance of the unsupervised learning is inferior to that of the supervised learning. In order to overcome this problem, considering the unsupervised learning and supervised learning together for anomaly detection, the present invention discloses the deep learning device and method for anomaly detection learning, which combine advantages of the unsupervised learning which does not require labeling and the supervised learning with high accuracy. 이상 감지 분야에서는 데이터 라벨링의 어려움과 데이터셋 불균형 때문에 널리 통용되는 지도학습보다는 비지도학습을 사용하게 된다. 그러나 비지도학습은 라벨링이 필요없지만 성능이 지도학습에 비해 떨어진다. 이러한 문제를 극복하고자 본 발명은 이상 감지를 위해 비지도학습과 지도학습을 함께 고려하여, 라벨링이 필요없는 비지도학습의 장점과 정확도가 높은 지도학습의 장점을 종합하는 이상 감지 학습을 위한 딥러닝 장치 및 방법을 제안한다.