Introduction

In this lecture and the next lecture, we’ll fully introduce how to actually train a neural network from scratch in practice. This lecture covers


Content

Activation Function

This section we discuss the pros and cons of different activation functions

Data Preprocessing

This section discuss different ways in preprocessing input data

Weight Initialization

Bad weight initialization can cause hard optimization. This section introduces some evident way of initializing

Regularization

Aside from L1 and L2 regularization, there are still other way of regularizing In training process, regularization add randomness to the training process while in testing process, we average out randomness