What Is Machine Learning?

The Spam Filter Story

Back in the early 2000s, email companies hired engineers to write rules for spam filters. The rules looked like this: "If the email contains 'FREE MONEY', mark it as spam." Or "If the sender is unknown and the subject is in ALL CAPS, reject it." Engineers would sit in meetings and debate which words were suspicious enough to block.

Then spammers adapted. They started writing "FR33 M0N3Y" or splitting words with spaces: "F R E E." Every time engineers wrote a new rule, spammers found a workaround. It was an endless arms race — and the humans were losing.

Then machine learning changed everything. Instead of humans writing rules, ML spam filters were trained on millions of examples of spam and legitimate email. The algorithm figured out its own patterns — ones no human would have thought of. Today's spam filters catch over 99% of junk mail. No human wrote those rules. The machine learned them.

That is machine learning: algorithms that improve automatically through experience.

The Official Definition

Computer scientist Tom Mitchell gave the most widely cited definition in 1997:

"A computer program is said to learn from experience E with respect to some task T and performance measure P, if its performance at task T, as measured by P, improves with experience E."

In plain English: a program gets better at a job the more it practices it.

For a spam filter:

• Task T: classify emails as spam or not spam
• Experience E: millions of labeled example emails
• Performance P: percentage of emails correctly classified

ML vs Traditional Programming

The key difference between ML and traditional programming is where the logic comes from.

Traditional programming:

Input data + Human-written rules → Output

A developer writes IF-THEN logic. The computer follows it exactly.

Machine learning:

Input data + Correct outputs → Machine learns the rules

You show the machine examples. It figures out the pattern.

This is why ML shines when the rules are too complex for humans to write explicitly — recognizing faces in photos, understanding spoken language, predicting stock movements.

The Three Types of Machine Learning

1. Supervised Learning

You provide labeled training data: every input has a known correct output. The model learns to map inputs to outputs.

Examples:

• Email spam detection (spam / not spam labels)
• House price prediction (price is the label)
• Image recognition (cat / dog labels)

Think of it as learning with a teacher who marks every answer right or wrong.

2. Unsupervised Learning

No labels. The algorithm must find structure on its own.

Examples:

• Customer segmentation: group buyers by purchase behavior without knowing the groups in advance
• Topic modeling: discover themes across thousands of news articles
• Anomaly detection: find unusual credit card transactions

Think of it as organizing a library where no one tells you the categories — you figure out a logical arrangement yourself.

3. Reinforcement Learning

An agent takes actions in an environment and receives rewards or penalties. Over time it learns the strategy that maximizes total reward.

Examples:

• AlphaGo beating world champions at Go
• Robots learning to walk
• Self-driving cars navigating intersections

Think of it as training a dog: reward good behavior, discourage bad behavior, repeat until the behavior is reliable.

Real-World Applications

Why Is ML So Powerful Right Now?

Three forces converged in the 2010s to make modern ML possible:

1. Big Data The internet generates 2.5 quintillion bytes of data every day. Social media, sensors, transactions, and searches all produce labeled or unlabeled examples that algorithms can learn from.

2. GPU Compute Graphics processing units, originally built for video games, turned out to be perfect for the matrix math inside neural networks. A GPU can perform thousands of parallel calculations simultaneously. Training that once took weeks now takes hours.

3. Better Algorithms Researchers developed techniques like deep learning, dropout regularization, and batch normalization that made training large models stable and accurate. Open-source libraries like scikit-learn, TensorFlow, and PyTorch put these tools in every developer's hands.

The ML Workflow at a Glance

Every ML project follows roughly the same path:

Raw Data
    |
    v
Data Cleaning & Preprocessing
    |
    v
Feature Engineering
    |
    v
Choose & Train Model
    |
    v
Evaluate Performance
    |
    v
Deploy & Monitor

Each step matters. A brilliant algorithm on bad data will fail. Clean data with the wrong model will underperform. The best ML practitioners spend more time on the first two steps than on model selection.

Key Takeaways

• Machine learning is programming by example, not by rules
• The three paradigms are supervised, unsupervised, and reinforcement learning
• ML thrives because big data, fast hardware, and mature algorithms arrived at the same time
• Every ML project follows a repeatable workflow from raw data to deployed model

The rest of this course walks through each step of that workflow in detail — starting with how to set up your environment and run your first complete ML program.