What is Machine Learning?

Improving performance on some task with experience

Experience: data

Tasks: lots of them

Performance: some measure of how correct you are

Basically: train a model on the data and see how well it works

Paradigms

• supervised learning
• unsupervised learning
• semi-supervised learning
• reinforcement learning
• active learning
• probably some others I'm missing

We'll focus on supervised learning.

Supervised Learning

Given training examples $(x, y)$, train a model $f$ to predict a $y$ given some $x$.

If $y$ is a label or category, this is called classification.

• e.g. $x$ is words in a document, $y$ is a topic (science, news, business, etc)
• e.g. $x$ is an image of a cell, $y$ is healthy or anemic cell

If $y$ is a real number, this is called regression.

• e.g. $x$ is a time series, $y$ is a stock price
• e.g. $x$ is data about a person, $y$ is how much to charge insurance premium

If $y$ is a rank or ordered list, this is called learning to rank.

• e.g. $x$ is a query, $y$ is a list of search results (homework 2)

If $y$ is a sequence, this is called sequence to sequence

• e.g. $x$ is a Chinese sentence, $y$ is an English sentence
• e.g. $x$ is speech signals, $y$ is a transcription of the speech