• Working with Flux Model Zoo
• Paying attention to the CNN model skipping the data preparation and training code

CNN explained

• ANN has some type of specialization of pattern detection

• Basis of CNN is convolutional layers

  • receives input, transforms input, ands outputs the transformed inputs to next layer
  • Convolutional layers need to specify the filters the layers have
    • filters are what detects patterns
      • patterns can be edge detectors, corners, circles, squares, etc.
      • detect objects like fur nose tail eyes mouth
        • detect class of animals like dog cat bird
  • A filter can be thought of as a matrix for which we decide the number of rows and number of columns the matrix has
    • values within the matrix are intialized with random numbers
  • The filter will slide over each 3x3 set of pixels from the input itself until it slid over every 3x3 block of pixels from the entire image
    • this sliding is called convolving.

Zero Padding in CNNs explained

• the convolved values actually reduce the dimensions of the input image

  n x n image f x f filter output size = (n - f + 1) x (n - f + 1)

  e.g. n=4 f=3 output size = 4-3+1 = 2 x 2

• the image is convolved it will get smaller and smaller

Zero Padding allows us to preserve the original input size specified on a per convolutional layer basis

• we add a border of pixels with the values of zeros around the image

valid padding means no padding, will decrease output shape same padding means input size is same as output size

from keras.models import Sequential
model = Sequential([
    Dense(16, activation='relu', input_shape=(20,20,3)),
    Conv2D(32, kernel_size=(3,3), activation='relu', padding='valid'),
    Conv2D(64, kernel_size=(5,5), activation='relu', padding='valid'),
    Conv2D(128, kernel_size=(7,7), activation='relu', padding='valid'),
    Flatten(),
    Dense(2, activation='softmax'),
])

model = Sequential([
    Dense(16, activation='relu', input_shape=(20,20,3)),
    Conv2D(32, kernel_size=(3,3), activation='relu', padding='same'),
    Conv2D(64, kernel_size=(5,5), activation='relu', padding='same'),
    Conv2D(128, kernel_size=(7,7), activation='relu', padding='same'),
    Flatten(),
    Dense(2, activation='softmax'),
])

Max Pooling in CNNs explained

Max Pooling is a type of operation added to following convolutional layers

• When added,reduces the dimensionality of images by reducing the number of pixels in the output from the previous layer input

  • we define a n x n region as a corresponding filter the max pooling operation

    • we then define a stride, how many pixels do we want our filter to move as it slides across the image

      Think of the n x n regions/blocks as "pools" of numbers, since were taking the max value of each pool… you get it.

• the network will be looking at larger areas, which reduces the amount of parameters in the network and reduces computational load

  • will also help to reduce overfitting

• Average pooling: take the average instead of the max (not used)

2….

Batch size in CNNs explained

The batch size is the number of samples that will be passed through to the network at one.

• an epoch is one single pass all the way through a network

  larger batch size = faster training

  • quality of model may degrade
    • test and tune batch size