Coursera Deep Learning Course4 Week2

Resnets

The identity block

The convolutional block (you can use this type of block when the input and output dimensions don ' t match up. The conv2d layer in the shortcut path was used to resize the input xx to a different dimension, so that the dimensions MATC H up in the final addition needed to add the value of the shortcut to the main path. (this plays a similar role as the Matrix Wsws discussed in lecture.) For example, to reduce the activation dimensions's height and width by a factor of 2, you can use a 1x1 convolution with a Stride of 2. The conv2d layer on the shortcut path does is not a use of any non-linear activation function. Its main role is-just apply a (learned) linear function that reduces the dimension of the input, so the the dimensions Match up for the later addition step. )

1x1 conv&inception Network Motivation
Stack cores of all sizes and volume that have been processed by the pooling layer are stacked together.

Use 1x1 conv to reduce computational capacity

Transfer Learning
It's about the same as before. If the data is small, you can replace the last layer only. The amount of data can be used to train the previous. The frame has settings that freeze the parameters of the front layer.

Data Augmentation
1). Mirroring
2). Random cropping
3). Rotation
4). Shearing
5). Local warping
6). Color Shifting
7). PCA

Data vs. hand-engineering
Two sources of knowledge can improve the effect of the model from both aspects, the more data less areas need manual design, such as features, structures, algorithms and so on.