ML M1: Introduction

What is Machine Learning?

Definition

Machine Learning (ML) is the process of building intelligent artifacts which learn to improve their performance at some task with experience.

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.

• Tom Mitchell, Machine Learning (1997)

ML uses statistical modeling to mathematically represent the task at hand. One of the most fundamental problems in ML is model generalizability, which refers to the gap in model performance on training data versus unseen instances (test data).

Subdivisions

ML is divided into three primary subfields:

• Supervised Learning: function approximation. Learning involves identifying a mapping between inputs $X$ and output(s) $y$.
  • SL is about induction - creating a general rule from a set of specific instances.
  \[f: X\rightarrow y\]

• Unsupervised Learning: data description. Learning involves extracting characteristics from input data $X$ to gain information relative to some general task (e.g., clustering, dimensionality reduction).

  \[f : X\]

• Reinforcement Learning: reward maximization in the context of an agent-environment dynamic. Learning involves identifying a policy, which maps environmental states to agent decisions.

  \[f : s \rightarrow a\]

Despite these differences, certain high-level concepts are shared across fields. Each field uses optimization as a means for learning - we define some objective function to evaluate model performance, then improve our model by optimizing the objective function. Furthermore, all ML problems are fully reliant on data. Since models are trained on data, their performance is highly dependent on data quantity and quality!