NEURAL NETWORK HARDWARE

The present invention relates to a neural network hardware. The neural network hardware according to an embodiment of the present invention comprises a label neuron part having multiple classes, wherein one class has at least two label neurons. Expression results of the label neurons are selected on...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: YI WOO SEOK, KIM JAE JOON
Format: Patent
Sprache:eng ; kor
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The present invention relates to a neural network hardware. The neural network hardware according to an embodiment of the present invention comprises a label neuron part having multiple classes, wherein one class has at least two label neurons. Expression results of the label neurons are selected one or more inference results performed through a neural network. 본 실시예에 의한 뉴럴 네트워크 하드웨어는 복수의 클래스를 가지며, 클래스 당 두 개 이상의 라벨 뉴런(label neuron)을 가지는 라벨 뉴런부를 포함하며, 상기 라벨 뉴런들의 발현 결과가 상기 뉴럴 네트워크를 통하여 수행된 1 회 이상의 추론 결과로 선택된다.