Introduction to Learning
Learning is the process of converting experience into expertise or knowledge.
For a machine the input to a learning algorithms is training data (representing experience) and the output is some expertise, which usually takes the form of another computer program that can perform some task.
Types of Learning
Supervised Learning
The learner is presented with examples:
\[(x_1,L(x_1)),\ldots,(x_n,L(x_n))\in \mathcal{X}\times\mathcal{Y}\]This is to say that it contains a list of pairs where the first element is data and the second is a provided label.
of labelled data $x_i$ $(L(x_i)$ is the label of $x_i$) and the task is to generalise the examples to a function $f:\mathcal{X}\rightarrow\mathcal{Y}$. If $\mathcal{Y}$ is finite, then $f$ is also called a classifier.
Unsupervised Learning
The learner aims to find structure (also called patterns) in data without using examples. Clustering data into similar ones is a typical unsupervised learning problem.
Reinforcement Learning
Learning from rewards or punishments in a dynamic setting in which an agent has a long-term goal (winning a game, driving a car). Different from supervised learning as actions of the learner are not directly classified.
Examples
Hand-Written Digit Recognition
Say that you are given a set of handwritten numbers. There are a set of labelled images and the rest are unlabelled.
An input image of 28 by 28 pixels is given as a vector $\vec x=(x_1,\ldots,x_{28\times28}$ of real numbers. Learn a classifier:
$f:\{\text{images}\}\rightarrow\{0,1,2,3,4,5,6,7,8,9\}$
Face Detection
Learn a classifier:
$f:\{\text{image windows}\}\mapsto \{\text{non-face, frontal face,}$ $\text{profile-face}\}$
Again a supervised learning problem:
- Start with training data:
- Image windows labelled with $\{\text{non-face, frontal face, profile-face}\}$.
- Generalise the training data to classifier $f$.
Spam Detection
Detect whether incoming email is pam or not.
A supervised learning problem:
-
Learn classifier:
\[f:\{\text{emails}\}\rightarrow\{\text{spam, not-spam}\}\]from labelled input data (emails).
Input data is word-count (e.g. word viagra in email indicated spam.)
Requires a learning system as “enemy” keeps innovating.
When do we Need Machine Learning?
- The problem’s complexity.
- The need for adaptivity.
Tasks that are too Complex to Program
- Tasks performed by animals/humans.
- There are numerous tasks that we human beings perform routinely, yet our introspection concerning how we do them is not sufficiently elaborate to extract a well defined program.
- Examples are given above.
- Machine learning programs achieve quite satisfactory results.
- Tasks beyond human capabilities.
- Another wide family of tasks that benefit from machine learning techniques are related to the analysis of very large and complex data sets.
- Examples include: astronomical data, weather prediction, web search engines.
- Meaningful information is buried in data archives that are way too large and too complex for humans to make sense of.
Adaptivity
- One limiting feature of programmed tools is their rigidity. Once a program has been written down and installed, it stays unchanged. However many tasks change over time or from one user to another.
- Machine learning tools (programs whose behaviour adapts to their input data) offer a solution to such issues; they are, by nature, adaptive to changes in the environment they interact with.
- Examples include individual handwritten text, spam detection, speach recognition.