Intro

Página 1

What is machine learning?

science of making computers learn without being explicitly programmed;

Tom Mitchell; : learning from experience E , to perform a Task T, on performance measure P , so long as performance P on T improves with experience E

Example ; playing checkers :

E - playing 10s of thousands of games against itself

T - playing checkers

P - probabilty of winning %

Example ; classifying email spam / not a spam

T - classify an email as spam/no spam

E - watching you classify email as spam/no spam

P - % of emails correctly classified spam/ no spam

Applications:

self-customizing programs : Amamzon , Netflix

Database Mining : of Large datasets

applications that cannot be programmed by hand :

handwriting recognition ; autonomous helicopter;

Página 2

Types of M/c Learning

Supervised | Unsupervised

Supervised : you provide algorithms with a dataset in which right answers were given;

ex: housing slave price vs Sq.ft

Data plot : Y => Price ; X => Square feet;

use linear regression to find the price for a given sq.ft

ex., Tumor => Malignant (1) | Benign (0)

classification problem : Logistic regression (Y => 1 or 0)

Página 3

Unsupervised learning:

Data sets are provided; Algorithms to define the classification;

example : customer data => algorithm would cluster them into different market segments

example : google news => clusters to collect all related news items being searched for : ex., Gulf Oil Spill

in supervised learning, right data is given to begin with to enable new classifications; (spam / no spam ; malign / benign)

in unsupervised learning => only a data is provided , no classification provided