APPARATUS AND METHOD FOR COMPRESSION OF NEURAL NETWORK MODEL

Disclosed are a device and method for lightening of a neural network model. The present invention can selectively apply, by a user, a model lightening method, and enable model lightening learning and inference to be performed without an expert knowledge in model lightening, thereby enabling to be pr...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: YOON CHANG OH, KIM DAE HYUN, IM JUN KYU, JI SUNG YOUNG, MIN JI UNG, AHN DUNG VO, LEE SI YOON, JUNG MIN SUNG, LEE WON BEEN
Format: Patent
Sprache:eng ; kor
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Disclosed are a device and method for lightening of a neural network model. The present invention can selectively apply, by a user, a model lightening method, and enable model lightening learning and inference to be performed without an expert knowledge in model lightening, thereby enabling to be provided by visualizing the size, the performance ratio, the calculation amount, the number of parameters, the usage amount of GPU or CPU memory during inference, and the improvement of inference speed and the result thereof according to model lightening. 신경망 모델의 경량화 장치 및 방법을 개시한다. 본 발명은 사용자가 모델 경량화 방법을 선택적으로 적용할 수 있고, 모델 경량화에 대한 전문지식 없이도 모델 경량화 학습과 추론을 수행할 수 있으며, 모델 경량화에 따른 모델의 크기, 성능비, 계산량, 파라미터수, 추론시 GPU 또는 CPU 메모리의 사용량 및 추론 속도의 개선과 그 결과를 시각화하여 제공할 수 있다.