00 - Intro to ML - lec. 1

ucla | CS M146 | 2023-04-03T14:42

---

Table of Contents

• Definitions
• Big Ideas
  • Types of Learning
    • Supervised
    • Unsupervised
    • Reinforcement
  • Machine Learning Goal
• Resources

Definitions

---

Big Ideas

---

Types of Learning

Supervised

• used for prediction
• given a labeled training set → learn a pattern/mapping for test set
• Regression → labels are continuous
• Classification → labels are discrete
• uses features/attributes to train on: tumor size, age, etc.

Unsupervised

• for detecting patterns
• given unlabeled data → find patterns (intrinsic structure)

Reinforcement

• used to interact in an environment
• not input data, just the environment → trial-and-error
• usually given a utility function (goal) → rewards the model for correct interactions

Machine Learning Goal

Given:

• a training dataset consists of $n$ labelled training instances for each has
• input features/attributes: $\overrightarrow{X}=[{\overrightarrow{x}}^{(1)},\dots ,{\overrightarrow{x}}^{(n)}]$ where $x^{(i)}\in {\text{R}}^d$ for $d$ features
• labels $\overrightarrow{y}=[y^{(1)},\dots ,y^{(n)}]$ where $y^{(i)}\in \text{R}$ or f we know limits (e.g. price >0) ${\text{R}}^{+}$

Output:

• hypothesis function: $h:{\text{R}}^d\to \text{R}$ from a hypothesis class (a family of functions e.g. for linear regression all linear functions) $h\in \mathcal{H}$ s.t. $h(\overrightarrow{x})\approx y$

Steps for learning $h$:

1. Represent (find) hypothesis class $\mathcal{H}$
2. Define a loss (function)
3. Optimize for the best $h\in \mathcal{H}$ s.t. $h(\overrightarrow{x})\approx y$

Resources

---

📌

**SUMMARY
**