GAN DATA PREDICTION METHOD BASED ON GENERATIVE ADVERSARIAL NETWORK AND APPARATUS IMPLEMENTING THE SAME METHOD
A method performed by a computing device according to an embodiment of the present invention comprises the steps of: pre-training a generator neural network based on an auto-encoder structure including an encoder and a decoder; and training a generative adversarial network (GAN) including a discrimi...
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: | A method performed by a computing device according to an embodiment of the present invention comprises the steps of: pre-training a generator neural network based on an auto-encoder structure including an encoder and a decoder; and training a generative adversarial network (GAN) including a discriminator neural network and the pre-trained generator neural network using a transfer learning method. The step of training a GAN using a transfer learning method includes the step of training the GAN with a decoder neural network of the generator neural network being fixed in an initial pre-trained state.
본 발명의 일 실시예에 따른 컴퓨팅 장치에 의해 수행되는 방법은, 인코더 및 디코더를 포함하는 오토인코더(auto encoder) 구조를 기반으로 하는 생성자(Generator) 신경망을 사전 학습(pre-training) 시키는 단계, 및 구분자(Discriminator) 신경망과 사전 학습된 생성자 신경망을 포함하여 구성되는 생성적 대립 신경망(GAN)을 전이 학습(transfer learning) 방식으로 학습시키는 단계를 포함하되, 생성적 대립 신경망(GAN)을 전이 학습(transfer learning) 방식으로 학습시키는 단계는, 생성자 신경망의 디코더 신경망을 사전 학습된 초기 상태로 고정한 상태에서 생성적 대립 신경망(GAN)을 학습시키는 단계를 포함한다. |
---|