Introduction to Machine Learning

• Machine learning constructs computer programs that automatically improve with experience.

What Is Machine Learning?

• The science of programming computers so they can learn from data.
• The ability of computers to learn without explicit programming.
• A computer program learns from experience E, regarding tasks T, measured by performance P, if its performance at tasks in T, as measured by P, improves with experience E.

Different Types of Machine Learning

• Supervised learning.
• Unsupervised learning.
• Reinforcement learning.

Supervised Learning

• Supervised learning involves a data set where the correct output is known, indicating a relationship between input and output.
• A training set of $N$ input-output pairs $\mathcal{D} = {(x_i, y_i)}^N_{i=1}$ is given
• The goal is to learn a function $f : \mathcal{X} \rightarrow \mathcal{Y}$ such that $y_i \approx f(x_i) \quad \forall i$
• Classification (pattern recognition) produces a discrete-valued output.
• Regression produces a real-valued output.

Example of Supervised Learning

• Distinguishing emails as spam using a training set of labeled emails.

Unsupervised Learning

• Unsupervised learning approaches problems with little to no prior knowledge of the results.
• Deriving structure from data without knowing the effect of variables.
• A training set of $N$ inputs $\mathcal{D} = {x_i}^N_{i=1}$ is given.
• The goal is to "find interesting structure" in the data.

Examples of Unsupervised Learning

• Clustering analysis: organizing objects into groups based on similarity.
• Density estimation: determining the distribution of data within the input space.
• Dimensionality reduction: projecting data from a high-dimensional to a low-dimensional space.

Reinforcement Learning

• Reinforcement Learning is inspired by how animals learn to achieve goals (also known as optimal control).
• The algorithm interacts with an environment, performs actions, and receives observations and rewards.
• The goal of the learner is to maximize the expected reward by learning a policy for taking actions based on observations.
• The policy is a mapping from observations to actions.
• The environment is typically modeled as a Markov Decision Process (MDP).

Examples of Machine Learning

• Image classification.
• Document classification and spam filtering.
• Predicting diabetes.
• Image completion.
• Speech recognition.

History of Machine Learning

• Marr's levels of analysis (1970s):
  • Computational level: the goal of the computation.
  • Algorithmic level: the specific algorithm used.
  • Implementation level: how to implement it in the brain or a computer program.
• AI effect:
  • Useful technology stops being called "AI".
  • "AI is whatever hasn't been done yet".

"What is not Machine Learning"

• Data mining:
  • Automatically discovering useful information in large data repositories.
  • Using machine learning as a "black box" to find curious correlations.
• Expert systems: extracting and encoding expert knowledge as rules.
• Hand-crafted rules: using if-then rules based on explicit criteria.

The Contents of the Book

• Part I: Foundations includes introduction, Bayesian decision theory, linear regression, classification, kernel methods, model assessment, and selection.
• Part II: Probabilistic models covers directed/undirected graphical models, exact/approximate inference, learning with latent variables, hidden Markov models, and state-space models.
• Part III: Advanced topics includes K-means clustering, mixture models, dimensionality reduction, advanced classification, combining classifiers, non-parametric models, and reinforcement learning, and how to read a paper.

Mathematical Notation

• $x$: Scalar random variable.
• $\mathbf{x}$: Vector of random variables.
• $\mathbf{X}$: Matrix of random variables.
• $x_i$: The $i$'th element of a vector.
• $\mathbf{x}_i$: The $i$'th row of a matrix.
• $\mathbf{X}^T$: Transpose of a matrix.
• $p(x)$: Probability mass function (PMF) if $x$ is discrete, or probability density function (PDF) if $x$ is continuous.
• $p(x \mid y)$: Conditional probability.
• $\mathcal{N}(\mu, \sigma^2)$: Normal (Gaussian) distribution with mean $\mu$ and variance $\sigma^2$.

Prerequisites

• Requires familiarity with basic probability theory, statistics, linear algebra, and calculus.