Convolutional Neural Network

Question 1

Basic overview of steps:

Answer 1

*Image as input
*Kernal to change each pixel

Question 2

How does the # of filters affect dimensions of image?

Answer 2

Alter third dimention

Question 3

Single math operation of 3x3 kernal with depth x?

Answer 3

sum(i=1 -> x33)(p_i w_i), p = pixel, w = weight

Question 4

What determines the dimension of output? 2*

Answer 4

*Stride
*Filters

Question 5

Kernals get a range of pixels into one pixel. What pixels do they extract?

Answer 5

Pixels along the x y and z direction.

Question 6

After one filter, what are the dimentions of the output?

Answer 6

yxyx1

Question 7

Convolution Layers?

Answer 7

How many times we apply a convolution operation.

Question 8

Activation?

Answer 8

Modify output values by some whatever function.

Question 9

Pooling

Answer 9

Depending on the type of pooling, decreasing the size of a matrix to prevent overlearning.

Question 10

does max pooling change size of output?

Answer 10

yes

Question 11

Dropout?

Answer 11

randomly sets some nodes to 0 to force neural network to train on more robust features. it ultimately prevents overlearning

Question 12

L1 regularization?

Answer 12

Sets a penalty term, calculated as the sum of the absolute values of weights and multiplied by a regularization parameter. this controls the strength of penalty. Overall, this forces the weights to remain small to prevent overfitting.

Question 13

Penalty in machine learning?

Answer 13

A condition for the loss function to encourage the model to adapt to certain properties.

Question 14

Cost function in machine learning?

Answer 14

Any function we want to minimize during learning.

Question 15

How do you mathematically describe cost?

Answer 15

Usually a loss function added by penalty function(s).