Deep Learning

Question 1

Step function

Answer 1

Ystep =

1 if X >= 0
0 if X < 0

Question 2

Sign function

Answer 2

Ysign =

+1 if X >= 0
-1 if X < 0

Question 3

Sigmoid function

Answer 3

Ysigmoid =

1
—
1+e-x

( 1 if X >= 0
0 if X < 0)

Question 4

Linear function

Answer 4

Ylinear = X

Question 5

Tanh function

Answer 5

Ytanh =

ex - e-x
————
ex + e-x

( +1 if X >= 0
- 1 if X < 0

Question 6

Number of parameters formula

Answer 6

1st convo layer
f * f * nc * nf + nf

After 1st convo layer
f * f * pnf * nf + nf

Question 7

Output size formula

Answer 7

n + 2p - f n + 2p - f
———— + 1 * ————— + 1
s s

Question 8

Derivates rules

Answer 8

x3 + 2x + 3

x3 = 3x2 (bring power down and -1)
2x = 2 (remain number, x = 1)
3 = 0 (any constant on own becomes 0)

(Anything attached by * will remain as it is)

Question 9

Euclidean distance

Answer 9

Square root of (a - b)2 + ….

Question 10

Manhattan distance

Answer 10

Absolute value of a - b + ……
( |a-b| + …

Question 11

Neural network output

Answer 11

Activation function(Total input * weights - bias)

Question 12

Learning a perceptron

Answer 12

e(p) = Yd(p) - Y(p)

( if error (e(p) is positive, we need to increase perceptron output (Y(p)). If negative, decrease Y(p)

Question 13

Relu function

Answer 13

Y = max(0,x)

Question 14

Leaky ReLu

Answer 14

Y = max(0.1, x)