Introduction in Machine Learning - Personalizing

with ML

What is Machine Learning?

“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”

-- Tom Mitchell, Carnegie Mellon University

A Spam Filter Application

E = The experience of classifying mail as spam or not spam.

T = The task of classifying emails (by subject, spam - not spam,

sender, ….), in our case whether it’s spam or not.

P = The probability that a mail is spam.

So the goal in (nearly) all ML applications is...

For a function that predicts a quantity Y as , where epsilon is the random error we try to minimize the difference between the predicted and the real real value of Y:

Getting our hands dirty with Linear Regression

http://bit.ly/28Zv65R

Prediction vs Classification

Prediction: Quantitative answers, predict a value of a continuous variable (height, weight, mileage, Click-through Rate, …)

Classification: Qualitative answers, find what is the observation you are looking at (a dog or a cat, span or not spam, dog breed, ….)

Classifying animals by their specie

Lets try Logistic Regression

http://bit.ly/29hSr1P

What is a neuron?

How many of them we have in our heads?

Answer

~86 billions!

Source and interesting read: http://bit.ly/1VSF5MK

And the Homo Sapien’s answer: The artificial neuron!

http://bit.ly/28Y93ef

Now let’s write one!

A neural network with multiple levels of features (intermediate processing layers)

What is Deep Learning

We can also optimize the structure of every layer, the transfer function, the activation function φ, the thresholds θ and much more.

Deep Learning is not limited in number of layers

http://bit.ly/28ZiM49

It’s time to go deep

Use Keras to train a network that classifies your photos against another person’s.

Homework

If you want to learn more:

Thank you for your time and effort!