A DECODING METHOD IN AN ARTIFICIAL NEURAL NETWORK AND AN APPARATUS THEREOF

Provided are a method for processing a token with a low probability in an auto-regressive based sequence generation artificial neural network, and a device thereof. A decoder in an artificial neural network can predict the probabilities of candidates for an output token based on at least one input t...

Full description

Saved in:
Bibliographic Details
Main Author LEE, MIN JOONG
Format Patent
LanguageEnglish
Korean
Published 22.05.2020
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:Provided are a method for processing a token with a low probability in an auto-regressive based sequence generation artificial neural network, and a device thereof. A decoder in an artificial neural network can predict the probabilities of candidates for an output token based on at least one input token input so far, determine an output token among the candidates based on the probabilities, and select one of the output token and a predefined special token based on the probability of the output token to determine a next input token. 오토 리그레시브 기반의 시퀀스 생성 인공 신경망에서의 낮은 확률 토큰의 처리방법 및 장치가 제공된다. 인공신경망에서의 디코더는 현재까지 입력된 적어도 하나의 입력 토큰에 기초하여, 출력 토큰의 후보들의 확률들을 예측하고, 확률들에 기초하여, 후보들 중 출력 토큰을 결정하고, 출력 토큰의 확률에 기초하여 출력 토큰 및 미리 정의된 특수 토큰 중 어느 하나를 선택함으로써, 다음 입력 토큰을 결정할 수 있다.
Bibliography:Application Number: KR20180139787