14
CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

CS230:Lecture3ThemathematicsofdeeplearningBackpropagation,Initializations,Regularization

KianKatanforoosh

Page 2: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

I– BackpropagationII– InitializationsIII– Regularization

Page 3: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

I- Backpropagation

Page 4: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Problemstatement

Page 5: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

A– LogisticRegressionbackpropagationforonetrainingexample

Page 6: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

B– LogisticRegressionbackpropagationforabatchofmexamples

Page 7: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Question:Youhavetrainedananimalclassifier.Canyoutellwhatpartoftheinputledtothisprediction?

Page 8: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

II- Initializations

Page 9: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Problemstatement

Page 10: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Inclass,you’veseenthat:

Page 11: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Thegoal:

Page 12: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Let’sprovethat: !"# "[%&'] = !"# "[%] → !"# +[%] = 1-[%&']

Page 13: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

Let’sprovethat: !"# "[%&'] = !"# "[%] → !"# +[%] = 1-[%&']

!"# "[%] = !"# .[%] = !"# / +',1% "1%&'2 345

16'Checkpoint:

Page 14: CS230: Lecture 3 · CS230: Lecture 3 The mathematics of deep learning Backpropagation, Initializations, Regularization Kian Katanforoosh

We’veshownthatforeverylayerl: !"# "[%] = -[%&']!"# +[%] !"# "[%&']