[INTRO TO TENSORFLOW FOR DEEP LEARNING] CNN 기초

INTRO TO TENSORFLOW FOR DEEP LEARNING (TensorFlow tutorials, Udacity) : CNN 기초

 

Feature == inputs

Lables == outputs

 

모든 머신러닝 모델은 다음 두 카테고리에 해당한다.

 

• regression (회기) :

- a model that outputs a single value

- just predicting a number that is the best fit to the data. (not trying to classify anything)

 

• classification (분류) :

- a model that outputs a probability distribution across several categories

- the output(consisting of a set of classes) is always a probability distribution

 

 

Two Main Concepts of CNN

 

• Convolutions

: the process of applying a kernel to an image

커널(필터)의 크기가 3*3이므로, 커널의 중앙 픽셀과 input value의 초록색 픽셀을 맞추어 컨볼루션 연산을 수행하여 Convoluted Image 초록색 칸에 연산값을 대입한다.

 

보통 zero-padding 적용

 

• MaxPooling

: the process of reducing the size of an input image by summarizing regions. 가장 큰 값 찾기.

 

down-sampling. grid size와 stride 결정해주어야함.

 

example code of Fashion MNIST with CNNs using Colab: 

https://github.com/tensorflow/examples/blob/master/courses/udacity_intro_to_tensorflow_for_deep_learning/l04c01_image_classification_with_cnns.ipynb

GitHub - tensorflow/examples: TensorFlow examples