Machine Learning Fundamentals & AI Taxonomy: 10-Part Video Stories to Level Up Your AI Knowledge

Welcome to our Machine Learning Taxonomy video series, led by Guillermo Germade, AI/ML & Backend Developer at Azumo. 

This 10-part series will guide you through the fundamental building blocks of machine learning, from the basics of supervised and unsupervised learning to more advanced topics, including anomaly detection, reinforcement learning, and large language models (LLMs), in just a few minutes per episode. 

Whether you're new to ML or need a refresher, these videos are designed to clarify concepts, spark new ideas, and show how machine learning applies to real-world cases.

Foundations of Modern AI

Guillermo Germade opens the training by outlining the learning objectives. He mentions that the series is useful for both business professionals seeking clarity on how AI can accelerate their business and developers looking to bridge into the world of machine learning. 

He walks you through the basics of traditional machine learning, starting with supervised and unsupervised learning, and shows how these ideas have evolved into today's AI techniques.

The session is focused on practical examples, helping you see where machine learning can make a real difference. The goal is to give you easy-to-understand knowledge that you can start using right away.

‍

What Is Machine Learning? A 5-Minute Primer

To start, Guillermo explains machine learning in simple terms. He mentions that machine learning is not about strict rules but rather an algorithm that learns from the data it’s given, allowing it to improve over time.

He compares traditional programming with machine learning, highlighting how ML can learn from experience and uncover hidden patterns. Guillermo also gives a quick overview of how machine learning fits into the broader AI field and briefly touches on key concepts like supervised, unsupervised, and reinforcement learning. This segment is perfect for those without a technical background, providing a solid foundation before getting into more complex topics.

‍

Supervised vs. Unsupervised Learning -The Big Picture

Guillermo walks us through the two most common types of machine learning: supervised and unsupervised learning.

He explains how each method works with data, using easy-to-understand examples like predicting rental prices or evaluating loan applications to show when you need labeled data and when you don’t.

In this section, he explains how supervised learning uses past examples and their results to make predictions. On the other hand, unsupervised learning looks for hidden patterns in data that doesn’t have labels. Guillermo also touches on some real-world challenges, like how performance can vary and the need to experiment, helping viewers understand how choosing the right method can impact a project’s success.

‍

Supervised Learning: Algorithms & Examples

Guillermo breaks down the two main types of supervised learning: regression and classification. He explains how labeled data helps train models and why it’s important to know the difference between predicting numbers (like sales) and categories (like loan approval).

He covers performance metrics, decision boundaries, and shows how models use factors like age, income, or education to make predictions or decisions. 

To make it easy to understand, he uses real-world examples like predicting ice cream sales, forecasting the weather, and deciding whether to approve a loan. 

Plus, you’ll get practical tips on how to choose the best algorithm for your specific needs, with clear examples and visuals to guide you.

‍

Unsupervised Learning: Clustering & Insights

In this section, we’ll explore how unsupervised learning algorithms like K-Means, DBSCAN, and PCA help find hidden patterns in data, all without needing target predictions.

Guillermo looks at real-world uses like customer segmentation and fraud detection, showing how businesses can group users by behavior or spot unusual activity, like flagging suspicious credit card transactions. 

With examples like Nike’s personalized marketing and fraud prevention in finance, he explains how unsupervised models can find structure in raw data, helping businesses make better decisions and take action, even when there are no labeled outcomes.

‍

Anomaly Detection vs. Classification: Credit-Card Fraud Walkthrough

Using a real-world credit card fraud example, Guillermo explains the difference between classification and anomaly detection. He shows that while both methods can spot fraudulent transactions, the choice depends on whether you have labeled data.

Classification models need past data labeled as "fraud" or "normal," while anomaly detection works without labels, identifying unusual patterns that don't match normal behavior. 

Guillermo also discusses important points like data imbalance, the impact of false positives, and how selecting the right model is crucial when dealing with rare but costly events, like fraud. This section provides a practical guide to picking the right approach when precision and context are key.

‍

Structured vs. Unstructured Data: Deep Learning Explained

Guillermo shows how neural networks are making breakthroughs in processing images, audio, and text, types of data that traditional models find challenging. He explains the difference between structured data (like spreadsheets or SQL tables) and unstructured data (like photos, voice recordings, and free-form text), setting the stage for understanding deep learning.

He introduces key architectures, like CNNs for image processing, RNNs for sequence data, and transformers for language tasks. Guillermo highlights why these models are great for real-world problems that are too complex for simpler methods.

However, he also points out that deep learning isn’t always the best choice, especially when dealing with structured data, where traditional models often perform better.

‍

Quick Tour & Future Outlook of Reinforcement Learning

Get a high-level understanding of how reinforcement learning works with agents, rewards, and learning from experience. Guillermo explains this approach by comparing it to how we train children, rewarding good behavior and punishing bad decisions. He also discusses how reinforcement learning gained attention with major breakthroughs, such as Google DeepMind’s AlphaGo and AI that plays chess better than humans.

Unlike supervised learning, reinforcement learning teaches agents to learn from their environment through trial and error. Guillermo explains where this technique is making an impact today, from video game AI to early-stage robotics, and hints at its potential to power the next wave of autonomous systems, even though it’s still mostly experimental in the business world.

‍

Large Language Models (LLMs) and AGI Toward General Intelligence

Guillermo explains how transformer-based large language models (LLMs), like GPT, work. He describes how they turn huge amounts of text, like from websites and books, into mathematical representations that understand meaning, intent, and context. He talks about how these models have quickly surpassed traditional NLP methods because of their amazing ability to generate language, reason across different topics, and even spark creativity.

Using real-world examples, Guillermo dives into the big question: can scaling LLMs lead to Artificial General Intelligence (AGI)? He also touches on the ethical concerns and skepticism from AI experts, pointing out that while LLMs are being widely used, their potential to reach human-like intelligence is still up for debate.

‍

Computer Vision Essentials: Object Detection with Supervised Learning

Guillermo explains how convolutional neural networks (CNNs) help machines "see" and understand visual data. He walks through how images are broken down into pixels, which are turned into data that neural networks analyze to spot patterns, like identifying streets, people, and objects.

Using real-world examples like security cameras, self-driving cars, face recognition, and medical imaging, he covers key computer vision tasks such as object detection, image classification, and segmentation. You’ll also get a hands-on overview of how these models are trained with labeled datasets, including tips on data preparation, annotation, and evaluation. This section provides a practical foundation to kick off your own computer vision projects.

‍

To Sum up

Stay tuned for the next session in the series, where we'll delve into Model Evaluation, focusing on how to assess performance in both classification and regression tasks. You'll learn the key metrics, when to use them, and how to interpret results to improve your models.

‍