Introduction (1): AI & Machine Learning

Introduction

Welcome to our introductory course on machine learning at UC Irvine! My name is Professor Alexander Islur, and I created these lectures for our introductory courses for undergraduates (CS 178) and graduates (CS 273).

Machine learning is generally considered a subfield of artificial intelligence (AI). AI focuses on building intelligent agents, which are often characterized by various tasks such as playing games like chess, enabling autonomous behavior in vehicles, and managing multi-agent dynamics, such as in robotic soccer competitions.

While machine learning can be more specific, it is also broadly applied. Its focus is on making predictions or decisions that improve with experience. Fundamentally, it is a data analysis science that combines aspects of computer science and computational thinking with traditional mathematics, including statistics and optimization.

This course is designed as an introductory class emphasizing practical application. While some theories will be included, the primary goal is to understand the principles of effective machine learning and how to improve outputs.

Types of Learning Problems

Machine learning problems often involve solutions that are challenging to describe explicitly. For instance, consider face detection in camera autofocus systems. How do we accurately describe what makes a face? Creating a set of rules to identify facial features is complex. Similarly, predicting how much you would enjoy a movie on Netflix poses the question: what types of movies do you usually like? These tasks exemplify supervised learning, where we have training data labeled with the correct answers.

Supervised learning problems generally fall into two categories:

1. Classification Problems: These involve predicting discrete values. For example, determining if a square contains a face or whether an email is spam.
2. Regression Problems: These involve predicting real-valued numbers, such as estimating that you may rate a movie 3.8 stars.

In contrast, unsupervised learning refers to problems where no specific signal must be predicted. Instead, the objective is to comprehend the data’s structure and similarity. Often called data mining, these problems focus on exploring and understanding data relationships.

An example of unsupervised learning applied to Netflix data might analyze the similarity among movies based on ratings. In another instance, a computer may observe images of a hand moving and can automatically organize them without semantic understanding, simply recognizing similarity based on visual features.

Beyond these, other learning approaches exist, including:

• Semi-supervised Learning: A blend of supervised and unsupervised learning, where examples with and without correct answers are used.
• Reinforcement Learning: Involves learning through indirect feedback about quality instead of precise answers, with applications in robotics and sequential decision-making tasks like poker, where a player may not know the correct actions until the game concludes.

In summary, machine learning involves a computational study of data that includes various subproblems. We will begin by exploring supervised learning, focusing on prediction problems that utilize training data consisting of input features and their corresponding labeled outputs.

Keywords

• Machine Learning
• Artificial Intelligence
• Intelligent Agents
• Supervised Learning
• Classification Problems
• Regression Problems
• Unsupervised Learning
• Data Mining
• Semi-supervised Learning
• Reinforcement Learning

FAQ

Q1: What is machine learning?
A1: Machine learning is a subfield of artificial intelligence focused on making predictions or decisions based on data analysis that improves with experience.

Q2: What types of problems does machine learning address?
A2: Machine learning primarily tackles supervised learning (classification and regression problems) and unsupervised learning problems.

Q3: What is the difference between classification and regression?
A3: Classification involves predicting discrete values (like determining spam emails), while regression predicts continuous real-valued numbers (like movie ratings).

Q4: What is unsupervised learning?
A4: Unsupervised learning involves exploring and understanding data structure and similarity without explicit output signals.

Q5: What is reinforcement learning?
A5: Reinforcement learning is learning through indirect feedback, often involving sequential decision-making where the outcomes are evaluated based on relative quality rather than precise answers.