Yioop - PHP Search Engine

Nov 3 In-Class Exercise Thread.

Please post to this thread your solution to the Nov 2 In-Class Exercise Thread.

Best,

Chris

(Edited: 2021-11-08)

Please post to this thread your solution to the Nov 2 In-Class Exercise Thread. Best, Chris

-- Nov 2 In-Class Exercise Thread

Our First Layer was

Conv2D(64, (3,3), padding="same", input_shape=(64, 64, 1))

Input m: 64 * 64 = 4096

Output n: No. of filters * Output shape = 64 * (64 * 64) = 262144

Therefore, using Glorot and Bengio initialization

`W_(i,j) ~ U (-(6/(4096+262144))^0.5, (6/(4096+262144))^0.5)` `W_(i,j) ~ U (-0.00475, 0.00475)`

(Edited: 2021-11-03)

Our First Layer was Conv2D(64, (3,3), padding="same", input_shape=(64, 64, 1)) Input m: 64 * 64 = 4096 Output n: No. of filters * Output shape = 64 * (64 * 64) = 262144 Therefore, using Glorot and Bengio initialization @BT@W_(i,j) ~ U (-(6/(4096+262144))^0.5, (6/(4096+262144))^0.5)@BT@ @BT@W_(i,j) ~ U (-0.00475, 0.00475)@BT@

-- Nov 2 In-Class Exercise Thread

((resource:Screenshot 2021-11-03 165701.png|Resource Description for Screenshot 2021-11-03 165701.png))

-- Nov 2 In-Class Exercise Thread

((resource:WhatsApp Image 2021-11-03 at 4.56.55 PM.jpeg|Resource Description for WhatsApp Image 2021-11-03 at 4.56.55 PM.jpeg))

-- Nov 2 In-Class Exercise Thread

((resource:WhatsApp Image 2021-11-03 at 4.55.38 PM.jpeg|Resource Description for WhatsApp Image 2021-11-03 at 4.55.38 PM.jpeg))

-- Nov 2 In-Class Exercise Thread

Conv2D(32, kernel_size=(3, 3), input_shape=(64,64,3))
m=64 x 64 =4096
n= 64 x 64 x 32 = 131072
Wi,j~U(-√6/(m+n), √6/(m+n)
Wi,j~U(-√6/135168, √6/135168)
Wi,j~U(-0.0066, 0.0066)

(Edited: 2021-11-03)

Conv2D(32, kernel_size=(3, 3), input_shape=(64,64,3)) m=64 x 64 =4096 n= 64 x 64 x 32 = 131072 Wi,j~U(-√6/(m+n), √6/(m+n) Wi,j~U(-√6/135168, √6/135168) Wi,j~U(-0.0066, 0.0066)

-- Nov 2 In-Class Exercise Thread

((resource:7, Giorot.jpg|Resource Description for 7, Giorot.jpg))

-- Nov 2 In-Class Exercise Thread

((resource:20211103_171826.jpg|Resource Description for 20211103_171826.jpg))

-- Nov 2 In-Class Exercise Thread

model.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', input_shape=(64, 64, 1)))

Input and Output (m,n): m= 64*64= 4096 n= 64*64*32= 131072

Resource Description for WhatsApp Image 2021-11-03 at 5.06.06 PM.jpeg

model.add(Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_uniform', input_shape=(64, 64, 1))) Input and Output (m,n): m= 64*64= 4096 n= 64*64*32= 131072 ((resource:WhatsApp Image 2021-11-03 at 5.06.06 PM.jpeg|Resource Description for WhatsApp Image 2021-11-03 at 5.06.06 PM.jpeg))

-- Nov 2 In-Class Exercise Thread

First layer:

 Conv2D(64, (3,3), padding="same", input_shape=(64, 64, 1))

which gives m = 64 * 64 and n = 64 * 64 * 64

Resource Description for inclass cnn layer.jpg

First layer: Conv2D(64, (3,3), padding="same", input_shape=(64, 64, 1)) which gives m = 64 * 64 and n = 64 * 64 * 64 ((resource:inclass cnn layer.jpg|Resource Description for inclass cnn layer.jpg))