Little Bits of Artificial Intelligence

1. Basic Definitions

2. Analogy: Learning to Recognize Faces

Machine Learning Analogy:

Imagine a security guard trying to recognize employees:

• The guard is given a checklist of features like hair color, height, and glasses.
• The guard uses these features to identify people.
• This is manual feature engineering—the algorithm needs pre-selected features.

Example ML Use Case:

• Email Spam Detection using decision trees or logistic regression trained on manually selected features like word frequency.

Deep Learning Analogy:

Now imagine a facial recognition camera is installed:

• It automatically learns from thousands of images.
• It identifies people even if lighting, hairstyle, or angle changes.
• No one tells it which features to use—it learns features hierarchically.

Example DL Use Case:

• Face Recognition using Convolutional Neural Networks (CNNs).
• Autonomous driving—detecting pedestrians or traffic signs using deep networks.

3. Comparison Table

4. Real-Life Neural Network Use Case Comparisons

5. Technical Summary

• ML: If structured data is given, and we know the important features, ML algorithms like decision trees, logistic regression, SVMs, or KNN can do a great job.
• DL: If complex data (like an image of a cat) is given, and we don’t know how to describe its “features,” a deep neural network can figure that out for us — automatically.

6. When to Use What?

7. Story style explanation

Title: The Tale of Two Learners – Alex and Dee

Once upon a time in a small tech-savvy town, there lived two cousins—Alex and Dee. Both were brilliant in their own ways and were asked by the town mayor to build a system to identify animals from photos taken in the local zoo.

Alex – The Machine Learner

Alex was practical and analytical. He said, “To identify animals, I need to know their features.”
So, he visited the zoo and observed each animal carefully.

He made a checklist:

• Number of legs
• Tail shape
• Ear length
• Fur color
• Stripes or spots?

Then, Alex used a decision tree algorithm and fed these manually chosen features into it.
His system worked well, but only when photos were taken in good lighting, perfect angles, and the animals were not moving.

Strengths of Alex’s Method (Machine Learning):

• Faster to build
• Easy to explain
• Works well with small structured data

Limitations:

• Struggled when a zebra was standing sideways (stripes were less visible)
• Failed when the animal was partly hidden
• Needed help every time a new animal or scenario was added

Dee – The Deep Learner

Now came Dee, who believed in learning through experience, not instructions.

She said, “Give me all the zoo pictures you have—hundreds, thousands, blurry or not—and I’ll let my brain figure it out.”

So, Dee built a deep neural network with many layers. Her system looked at raw pixels of images and began learning:

• The first layer noticed simple edges and shapes
• The next layer learned how those shapes combined into animal parts
• Deeper layers began identifying full animals—a lion’s face, an elephant’s trunk, even when lighting was poor or animals were far away.

Over time, Dee’s system became incredibly accurate. It could even tell apart a Bengal tiger and a white tiger, which Alex’s system confused.

Strengths of Dee’s Method (Deep Learning):

• Learned features automatically from data
• Got better with more data
• Could handle complex scenarios like poor light, partial views

Limitations:

• Took a longer time to train
• Needed a powerful computer (GPU)
• Was like a black box—hard to understand how it made some decisions

Moral of the Story

Alex’s approach (Machine Learning) is like learning with a textbook—structured, clear, but limited to what’s explicitly taught.

Dee’s approach (Deep Learning) is like learning by experiencing the world—one might not know exactly how he learned it, but his brain starts recognizing patterns effortlessly.

Neural Network Context Recap

Next – Deep Learning Categories