Machine Learning Foundations

Course Content

Machine Learning Foundations

14 sections · 70 lessons

What is the difference between AI, ML, and Deep Learning?

Nested circles, drawn from outside inAI — any machine doing intelligent tasksML — learns patterns from dataDeep learning — many-layer neural networksGenerative AI — LLMs and image generators
Every LLM sits inside all four layers, but a rule-based delivery-partner assigner is AI that never touches the inner three.

What you need to know

Artificial intelligence: the goal

AI is any technique that makes a computer do something we would call intelligent: plan a route, play chess, answer a question. Older AI was often rule-based. A 1980s medical "expert system" was thousands of if-then rules written by doctors. It counts as AI, but it does not learn.

Machine learning: learning from data

ML is the part of AI where the system improves from examples. Classical ML algorithms such as linear regression, decision trees, random forests and gradient boosting work very well on tabular data, meaning rows and columns like a spreadsheet. They usually need feature engineering: a person decides which inputs matter and prepares them, for example turning a date of birth into age.

Deep learning: many-layer neural networks

Deep learning uses neural networks with many layers. Each layer transforms its input a little, and together they learn features by themselves. For images, early layers learn edges, later layers learn shapes like eyes or wheels. Nobody hand-codes "what an eye looks like".

How to choose between classical ML and deep learning

Classical ML

  • Best on tabular data (loans, churn, prices)
  • Works with thousands of rows
  • Trains in seconds on a laptop
  • Easier to explain to a risk team

Deep learning

  • Best on images, audio and text
  • Usually needs far more data
  • Needs GPUs for training
  • Learns features itself, harder to explain

For a telecom churn table with 30 columns, gradient boosting often beats a neural network and is far cheaper. For recognising a cheque signature from a photo, deep learning wins easily.

A real-life example

Take a food-delivery app.

  • AI but not ML: a rule that assigns the nearest free delivery partner to an order. It is automated decision-making with no learning.
  • ML but not deep learning: a gradient-boosting model that predicts delivery time from distance, restaurant preparation history, time of day and rain.
  • Deep learning: a neural network that reads the photo a customer uploads with a complaint and checks whether the food is actually spilled.
  • Generative AI (a part of deep learning): an LLM that drafts the apology message to the customer.

All four are "AI" in a press release. Only three learn from data, and only two are deep learning.

Follow-up questions to expect

  • "Where does generative AI fit?" — Inside deep learning. LLMs and image generators are large neural networks trained to produce new content.
  • "Why did deep learning take off after about 2012?" — Three things arrived together: large labelled datasets, GPUs fast enough to train big networks, and better training methods.
  • "Is a chess engine AI?" — Yes. A classic chess engine that searches moves with hand-tuned rules is AI without ML. Newer engines add learned neural networks to judge positions.