r g.ai ork - deep learningcs230.stanford.edu/files/c1m4.pdf · andrew ng applied deep learning is a...

deeplearning.ai

Deep NeuralNetworks

Deep L-layerNeural network

AndrewNg

What is a deep neural network?

logistic regression

1 hidden layer

2 hidden layers

5 hidden layers

AndrewNg

Deep neural network notation

deeplearning.ai

Deep NeuralNetworks

Forward Propagationin a Deep Network

AndrewNg

Forward propagation in a deep network

deeplearning.ai

Getting your matrixdimensions right

Deep Neural Networks

Andrew Ng

Parameters !["] and "["]

$%&!&"

Andrew Ng

Vectorized implementation

!"#!#"

deeplearning.ai

Whydeeprepresentations?

DeepNeuralNetworks

AndrewNg

Intuition about deep representation

!"

AndrewNg

Circuit theory and deep learningInformally: There are functions you can compute with a “small” L-layer deep neural network that shallower networks require exponentially more hidden units to compute.

deeplearning.ai

Building blocks ofdeep neural networks

Deep NeuralNetworks

AndrewNg

Forward and backward functions

AndrewNg

Forward and backward functions

deeplearning.ai

Parameters vsHyperparameters

Deep Neural Networks

Andrew Ng

What are hyperparameters?Parameters: ! " , % " ,! & , % & ,! ' , % ' …

Andrew Ng

Applied deep learning is a very empirical process

cost !

# of iterations

Idea

Experiment Code

deeplearning.ai

What does thishave to do with

the brain?

Deep Neural Networks

Andrew Ng

Forward and backward propagation!["] = #["]$ + &["]'["] = ( " (! " )![$] = #[$]'["] + &[$]'[$] = ( $ (! $ )

'[%] = ( % ! % = +,

…

…

-![%] = '[%] − +-#[%] = 1

0-! % ' % &

-&[%] = 1012. sum(d! % , 9:;< = 1, =>>2-;0< = ?@A>)

-![%'"] = -# % &-! % (( % (! %'" )

-#["] = 10-! " ' " &

-&["] = 1012. sum(d! " , 9:;< = 1, =>>2-;0< = ?@A>)

-!["] = -# % &-! $ (( " (! " )