Machine Learning

Machine Learning - field of study that gives computers the ability to learn without being explicitly programmed

A computer is said to learn from experience $E$ with respect to some task $T$ and some performance measure $P$

  • if it's performance of $T$,
  • as measured by $P$,
  • improves with experience $E$


E.g. email filtering:

  • $T$: classifying email as spam/not spam
  • $E$: watching you label as spam/not spam
  • $P$: the number of emails correctly classified as spam/not spam


Examples of Machine Learning:

  • db-mining
    • web-data clicks, etc
  • automation
    • autonomous helicopter
    • NLP
    • Computer Vision
  • self-customizing software
    • amazon, netflix, etc
  • understanding human learning


Supervised Learning

"right answers" are given

  • i.e. the algoritm takes a training set $\{(x^{(i)}, y^{(i)})\}$
  • and then predicts a value for / classifies a data example $x$


Regression

Regression - predict: continuous values

Examples:

  • You have a large inventory of identical items
  • You want to predict how many of these items will sell over the next 3 months


Main tool


Classification

Classification - assigning to a group (0, 1) etc: discrete values

$y \in \{0, 1\}$ - binary classification problem

  • 0 - negative class, connected with absence of smth (not spam)
  • 1 - positive class, connected with presence of smth (spam)

Tools:


Unsupervised Learning

  • just given the data, no labels
  • can we find a structure in the data?
  • we don't tell the algorithm what are the categories

i.e. we're given only $\{x^{(i)}\}$, no $y^{(i)}$s

applications

  • news segregation
  • social network analysis
  • clustering
  • market segregation

Clustering

The goal is to automatically group the data into coherent subsets (or clusters)



Sources