Distributed parallel deep learning system server and method
An objective of the present invention is to provide a distributed parallel deep-learning system and method that can perform deep-learning learning without collection. The distributed parallel deep learning system includes: a plurality of work servers that omni-directionally transmits a calculated we...
Gespeichert in:
Hauptverfasser: | , , |
---|---|
Format: | Patent |
Sprache: | eng ; kor |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | An objective of the present invention is to provide a distributed parallel deep-learning system and method that can perform deep-learning learning without collection. The distributed parallel deep learning system includes: a plurality of work servers that omni-directionally transmits a calculated weighted value by performing a learning process of a first layer by a learning model stored in advance based on raw data, performs the learning process of remaining layers by reflecting the weighted value transmitted from the plurality of work servers; and a learning management server that reversely transmits the calculated total weighted value by collecting the weighted value to each of the plurality of work servers, wherein each of the plurality of work servers applies a total weight propagated backwards to the first layer.
본 발명은 분산 병렬 딥러닝 시스템 및 방법에 관한 것으로, 분산 병렬 딥러닝 시스템은 로우데이터에 기초하여 미리 저장된 학습 모델에 의해 제1 레이어의 학습 과정을 수행함으로써 연산된 가중치를 전방향 전파하는 복수개의 작업서버; 및 상기 복수개의 작업서버로부터 전파되는 가중치를 반영하여 나머지 레이어의 학습 과정을 수행하고, 상기 가중치를 취합하여 연산한 총(total) 가중치를 상기 복수개의 작업서버 각각에 역방향 전파하는 학습관리서버를 포함하고, 상기 복수개의 작업서버 각각은 역방향 전파된 총(total) 가중치를 상기 제1 레이어에 적용하는 것을 특징으로 한다. |
---|