Introduction of Deep Learning

some introductions

• Machine Learning is close to Looking for a Function
• The model is a set of function
• a set of function -> goodness of functin F <- training data
• framework
  – get a model consists of a set of function
  – through training data’s training to get the goodness of function f
  – pick the best function f*
• three steps for deep learning
  – define a set of function(is also called neural network)
  – goodness of function
  – pick the best function
• Neural Network
  – different connections leads to differeent network structure
  – each neurons can have different values of weights and biases,weights and biases are network parameters
  
  
  – deep learning means many hiden layers,each node means a function consists of weights and bias
• output layer
  – the output of network can be any value and may not be easy to interpret
  
  – the value y is the probability of each output value and you need to decide the network structure to let a good function in your function set

training data

• preparing the training data:images and their labels.The learning target is defined on the training data
  
  
• a good function should make the loss of all examples as small as possible

pick the best function

• the target is to find network parameters weights* and bisa* that minimize the total loss value L
• how to find the best parameters:
  
  – different initial parameters->reach different minima,so different results,so how to choose the init parameters is important

Why Deep?

Deeper is Better?

• actualy,more parameters,better performance
• the thin+tall neural network is better than the fat+short.
• deep->modularization->less training data