The weight implies the strength of the connection across neurons
A neural networks can be viewed as a computational graph of elementary units, (such as least squares regression of logistic regression) where the connection of these units are combined in a structured way to enable a more complex function of the input data.
Two key advantages of NNs:
Provides a higher-level abstraction of expressing semantic insights about data domains by architectural design choices in the computational graph.
Adding or removing neurons from the achitecture provides a simple way to adjust the complexity of a model.
Conventionally, the first ConvLayer is responsible for capturing the low-level features such as edges, color, and gradient orientation.
Max pooling performs better than average pooling. It de-noises and reduces dimensionality. Average pooling only reduces dimensionality, maintainin noise.
The Convolutional Layer and Pooling Layer compine to form the i-th layer of a CNN.
Add a fully-connected layer (multi-level perceptron), flatten the image to a column, feed it into a feed-forward NN, and apply backpropagation to every iteration of training. Softmax classification layer at end.