AI.md

marp	paginate
true	true

AI: From the pages of science, to the future of our lives.

by Ahmadreza Anaami

---

Goals

In this lab you will:

• have a brief introduction on AI and ML
  • basic concept of machine learning
• main algorithm
  • supervised
    • regression
    • classification
  • unsupervised
• Learn to implement linear regression

---

---

machine learning

Subfield of AI in order to make intelligent machine

Field of study that gives the computers the ability to learn without explicitly programmed

• Arthur samuel 1959

---

Supervised learning

• learn from being given right answers
  • correct pairs of input (x) and output (y)

---

input	output	application
email	spam?(0,1)
audio	text transcript
English	Spanish
ad-User	click?(0,1)
image-radar	position of other car
image of phone	defect?(0,1)

---

Regression

predict a number from infinitely many possible outputs

---

Classification

many inputs some outputs called categories

Size	diagnosis
6	1
8	1
2	0
5	0
1	0
7	1
5.6	1
12	1
3.5	0

---

---

Recap

In every living man a child is hidden that wants to play

---

---

Unsupervised learning

Find something intreating in (unlabeled) Data

• Clustering
• anomaly detection
• Dimensionality reduction

---

---

---

---

linear regression

$$

f_{w,b}(x) = wx + b

$$

w , b > parameters , weight x > single feature

---

Cost function

$$J(w,b) = \frac{1}{2m} \sum\limits_{i = 0}^{m-1} (f_{w,b}(x^{(i)}) - y^{(i)})^2$$

where

$$yHat = f_{w,b}(x^{(i)}) = wx^{(i)} + b $$

m = number of example

---

Example

choose w to minimize J(w)

$$yHat = f_{w}(x^{(i)}) = wx^{(i)} $$

$$J(w) = \frac{1}{2m} \sum\limits_{i = 0}^{m-1} (f_{w}(x^{(i)}) - y^{(i)})^2$$

---

---

Gradient descent

a more systematic way to minimize j(w,b)

$$ \newline

\begin{align*} \text{repeat}&\text{ until convergence:} ; \lbrace \newline ; w &= w - \alpha \frac{\partial }{\partial w} J(w,b) ; \newline \newline b &= b - \alpha \frac{\partial }{\partial b} J(w,b) \newline \rbrace \end{align*}$$

---

Derivative of the Cost func

$$J(w,b) = \frac{1}{2m} \sum\limits_{i = 0}^{m-1} (f_{w,b}(x^{(i)}) - y^{(i)})^2$$

$$ \begin{align} \frac{\partial }{\partial w}J(w,b) &= \frac{1}{m} \sum\limits_{i = 0}^{m-1} (f_{w,b}(x^{(i)}) - y^{(i)})x^{(i)} \\ \frac{\partial }{\partial b}J(w,b) &= \frac{1}{m} \sum\limits_{i = 0}^{m-1} (f_{w,b}(x^{(i)}) - y^{(i)}) \\ \end{align} $$

---

optional

if alpha is too small :

if alpha is too big :

$$\alpha \frac{\partial }{\partial w} J(w,b)$$

calculate Derivative : ☻

---

Lets get our hands dirty

• step 1
  • read our data
• step 2
  • write our functions
    • calculate yHat
    • calculate cost
    • calculate gradient
    • implement gradient descent
• step 3
  • congratulations