week4

Page 1

if number of features (n) increase, combination of features dramatically increase : (n^3) :

so classification and linear regression may not work as it becomes computationally intensive;

training example : set of cars : not cars

computer recognizes them as pixels;

for 100 x 100 pixel : logistic regression including all the quadratic terms (xixj) as features, will result in train about how 0 million features

Page 2

Neural networks mimic the brain:

brain learns from sensors;

auditory cells can be trained to see and learn!

neurons! interconnect and communicate with each other!

Page 3

Layer 1 => Layer 2 => Layer 3

I/p => Hidden layer => Output Layer

x0

x1

x2 → [ ] → hθ(x)

Page 4

Example: If layer 1 has 2 input nodes and layer 2 has 4 activation nodes.

Dimension of Θ(1) is going to be 4×3 where sj=2 and sj+1=4, so sj+1×(sj+1)=4×3.

Page 5

a1(2)=g(z1(2))

a2(2)=g(z2(2))

a3(2)=g(z3(2))

zk(2)=Θk,0(1)x0+Θk,1(1)x1+⋯+Θk,n(1)xn

Page 6

intuitions:

neural networks can be implemented for AND | OR functions

x1, x2 } binary (0,1)

y = x1 AND x2

h-theta(x) = g (theta0 + theta1*x1 + theta2-x2)

theta0 = -30 ; theta1 = 10 ; theta2 = 10 ; represents AND gate

theta0 = -30 ; theta1 = 20 ; theta2 = 20 ; represents OR gate