Machine Learning

Front 1

What are some definitions of Machine Learning?

Back 1

Classic: Arthur Samuel described it as "the field of study that gives computers the ability to learn without being explicitly programmed."

Modern: Tom Michell described it as "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."

Front 2

Give an example of the modern definition of Machine Learning using:

E = the experience of playing many games of checkers

P = the probability that the program will win a game of checkers

T = the task of playing checkers

Back 2

Machine learning is defined by a program that with increased experience, E, improves their performance measure, P, at task T then it is said to be "learning."

Front 3

What is supervised learning?

Back 3

In supervised learning, we are given a data set with knowledge of the correct output, given some declared input.

Supervised learning is classified into two main categories:

1) Regression: map input into a continuous function

2) Classification: map input into a discrete function

Front 4

For the following examples, what type of machine learning problem is it?

1) Given data about the size of a house on the real estate market, try to predict their price.

2) Given data about the size of a house on the real estate market, determine whether the house will sell for more or less than it's asking price.

Back 4

1) Price changes continuously based upon the size of the house, so it is a regression type problem of supervised learning.

2) Since the output falls into discrete values of "more than asking price" or "less than asking price" it is a classification problem of supervised learning.

Front 5

What is unsupervised learning?

Back 5

Unsupervised learning is the approach to a problem with little or no idea of what our results should look like.

Front 6

Give two ways of describing unsupervised learning.

Back 6

1) Clustering: we group data together based upon relationships among the variables in the data

Example: group 1000 essays on US economics based on variables such as word count, sentence length, page count etc.

2) Associative Patterns: we store patterns or things that go together.

Example: a doctor, over many years of practice, associates certain patient characteristics to a given illness. There is no logical mapping, just a data set of conditions and illnesses that is not exhaustive.