[memo] cnn hyperparameter, neural network architecture

(23.01.22)

 

 

 

convolution neural network 사용할 때의 hyperparameter

 

1) convolution

- filter 크기

- filter 개수

 

2) pooling

- window 크기

- window stride

 

3) fully-connected

- layer의 개수

- neurons의 개수

 

 

 

neural network 아키텍쳐 종류

 

- one to one: vanila nerual networks

- one to many: image captioning (image -> sequence of words)

- many to one: sentiment classification (sequence of words -> sentiment(감정))

- many to many(1): machine translation (sequence of words -> sequence of words)

- many to many(2): video classification on frame level