최대 1 분 소요

Seq2Seq with Attention

Seq2Seq2 Model

  • Take a series of words as input to the Encoder and output a series of converted words to the Decoder
  • Latent vector storing information about the input from Encoder
  • Can handle variable-length inputs and outputs

Problem - Bottleneck problem

  • Fixed-size latent vectors can’t hold all of the long input information

How to solve Bottleneck problem

  • Attention
    • At each time step of the Decoder, if focuses on the part of the Hidden-state vector created from the input Sequence that it needs so that if can take that information and write it.

댓글남기기