ELECTRONIC DEVICE AND METHOD FOR CONTROLLING SAME

Disclosed are an electronic device and a method for controlling same. An electronic device of the present disclosure may comprise: a memory which stores a pre-trained neural network model and training data; and a processor which obtains a first loss function by using a label corresponding to the tra...

Full description

Saved in:
Bibliographic Details
Main Authors KAPOOR, Parichay, LIM, Geunsik, MOON, Jijoong, LEE, Jihoon, HAM, Myungjoo
Format Patent
LanguageEnglish
French
Korean
Published 07.07.2022
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:Disclosed are an electronic device and a method for controlling same. An electronic device of the present disclosure may comprise: a memory which stores a pre-trained neural network model and training data; and a processor which obtains a first loss function by using a label corresponding to the training data and output data obtained by inputting the training data into the neural network model, obtains the magnitude of the change amount of a weight of each of a plurality of layers included in the neural network model on the basis of the first loss function, and trains the neural network model by updating a weight of at least one layer in which the magnitude of the change amount of the weight exceeds a first threshold value, among the plurality of layers. L'invention divulgue un dispositif électronique et son procédé de commande. Un dispositif électronique selon la présente divulgation peut comprendre : une mémoire qui stocke un modèle de réseau neuronal préformé et des données d'apprentissage ; et un processeur qui obtient une première fonction de perte en utilisant une étiquette correspondant aux données d'apprentissage et aux données de sortie obtenues en entrant les données d'apprentissage dans le modèle de réseau neuronal, qui obtient l'amplitude de la quantité de changement d'un poids de chaque couche d'une pluralité de couches incluses dans le modèle de réseau neuronal sur la base de la première fonction de perte, et qui forme le modèle de réseau neuronal en mettant à jour un poids d'au moins une couche dans laquelle l'amplitude de la quantité de changement du poids dépasse une première valeur de seuil, parmi la pluralité de couches. 전자 장치 및 이의 제어 방법이 개시된다. 본 개시의 전자 장치는 기 학습된 신경망 모델 및 학습 데이터를 저장하는 메모리 및 학습 데이터를 신경망 모델에 입력하여 획득된 출력 데이터 및 학습 데이터에 대응되는 라벨을 이용하여 제1 손실 함수를 획득하고, 제1 손실 함수에 기초하여 신경망 모델에 포함된 복수의 레이어 각각의 가중치의 변화량 크기를 획득하고, 복수의 레이어 중 가중치의 변화량의 크기가 제1 임계값을 초과하는 적어도 하나의 레이어의 가중치를 업데이트하여 신경망 모델을 학습시키는 프로세서가 포함될 수 있다.
Bibliography:Application Number: WO2021KR15591