Table of contents
3. Encoder-decoder architecture
1. CNN angle
Taking the convolutional neural network as an example, the input is a cat, and the image category is output after feature extraction.
Therefore, it can be simply understood here that the feature extraction process is the encoder, and the classification process is the decoder.
Right now:
Encoder: Program the input into an intermediate expression form (feature) [feature extraction]
Decoder: Decodes the intermediate representation into output. 【Classifier】
2. RNN angle
Taking the cyclic neural network as an example, the input is a text, and the output is extracted after feature extraction.
therefore,
Encoder: Represent text as a vector [feature extraction]
Decoder: vector representation as output [classifier]
3. Encoder-decoder architecture
A model is divided into two pieces:
1. Encoder processing output
2. Decoder generates output
Using the encoder-decoder architecture model, the encoder is responsible for representing the input and the decoder is responsible for the output