Level 01: AI Awakening

What is Artificial Intelligence?

Imagine you're teaching your little brother to recognize dogs. At first, he might think every four-legged animal is a dog. But after you show him lots of pictures—"That's a dog," "That's a cat," "That's a dog"—he starts to notice patterns. Dogs have wet noses. Cats have pointy ears. Over time, he gets better and better at telling them apart.

That's exactly what AI does. Except instead of a little brother, we have a computer. And instead of showing it a few pictures, we show it millions. The computer looks for patterns—shapes, colors, textures—and slowly learns to recognize things on its own.

            Key Idea: AI is a computer program that learns from examples, just like humans learn from experience.
          

The Pattern Matching Game

Let's play a simple game. Below is a 5×5 grid. Click on cells to create a pattern, and watch how the "AI" tries to predict what kind of pattern you're making!

Introduction to Machine Learning

Artificial Intelligence (AI) refers to computer systems capable of performing tasks that typically require human intelligence. Machine Learning (ML), a subset of AI, enables systems to automatically learn and improve from experience without being explicitly programmed.

The fundamental mechanism underlying ML is pattern recognition. Given a dataset containing input-output pairs, the learning algorithm identifies underlying statistical patterns that map inputs to outputs. This learned mapping function can then generalize to previously unseen inputs.

            Definition: Machine learning is the study of computer algorithms that improve automatically through experience 
            and by the use of data.
          

Supervised Learning Paradigm

In supervised learning, the algorithm is presented with labeled training data—pairs of input features and corresponding target outputs. The objective is to learn a function that can predict the target output for new, unseen inputs.

Formally, given a training set D = {(x₁, y₁), (x₂, y₂), ..., (xₙ, yₙ)}, we seek to learn a function f: X → Y such that f(x) ≈ y for new examples.

🎮 Pattern Recognition Demo

Click cells to activate them. The AI will try to classify your pattern in real-time.

AI Predictions:

Horizontal

20%

Vertical

15%

Diagonal

10%

Random

55%

How Does a Computer Learn?

When you learn to ride a bike, you don't read a manual—you try, fall, adjust, and try again. Learning from mistakes is powerful!

Computers learn the same way. Here's the secret formula:

            Make a guess - The computer predicts the answer
Check the mistake - See how wrong it was
Adjust slightly - Change the approach
Repeat! - Do it thousands of times

          

The Guessing Game

Let's say we want to teach a computer to predict house prices. We show it:

House with 2 bedrooms → $200,000
House with 3 bedrooms → $300,000
House with 4 bedrooms → $400,000

The computer might start by guessing: "Maybe each bedroom adds $50,000?"

For the 2-bedroom house: Guess $100,000 ❌ (Actual: $200,000)
Mistake: $100,000 too low!
New guess: "Maybe $100,000 per bedroom?"

After testing this new guess on all the houses, it works! The computer has learned.

Learning as Optimization

Machine learning can be understood as an optimization problem. The goal is to find model parameters that minimize a loss function, which quantifies the discrepancy between predicted and actual outputs.

The Learning Algorithm

Given:

A model with parameters θ
A loss function L(θ; D) measuring error on dataset D
A learning rate η controlling step size

The learning process iteratively updates parameters:

Gradient Descent Update: θ t+1 = θ t - η \nabla θ L(θ t; D)

Where ∇_θL is the gradient of the loss with respect to parameters, indicating the direction of steepest increase in loss. By moving in the opposite direction, we minimize the loss.

Example: Linear Regression

For predicting house prices based on bedrooms:

Python

Pseudocode

# Simple linear regression
def predict(bedrooms, weight, bias):
    return weight * bedrooms + bias

# Training data
houses = [(2, 200000), (3, 300000), (4, 400000)]

# Initial parameters
weight = 50000  # Initial guess: $50k per bedroom
bias = 0

learning_rate = 0.01

# Training loop
for epoch in range(1000):
    total_error = 0
    for bedrooms, actual_price in houses:
        # Forward pass
        predicted = predict(bedrooms, weight, bias)
        
        # Calculate error
        error = predicted - actual_price
        total_error += error ** 2
        
        # Update parameters (gradient descent)
        weight -= learning_rate * error * bedrooms
        bias -= learning_rate * error
    
    if epoch % 100 == 0:
        print(f"Epoch {epoch}: Error = {total_error:.2f}")

print(f"\nLearned: ${weight:.0f} per bedroom + ${bias:.0f}")
          

Why Does This Matter?

You use AI every single day, probably without realizing it:

What You Do	The AI Behind It
Watch YouTube videos	AI learns what you like and recommends more
Use Snapchat filters	AI recognizes your face in real-time
Ask Siri or Alexa	AI understands your speech and responds
Play video games	AI controls the enemies and adapts to your skill
Use auto-correct	AI predicts what word you're typing

Where We're Headed

In this course, we're going to start from scratch and build up to understanding one of the most advanced AI research papers from 2025. By the end, you'll understand:

✅ How neural networks work (Level 2)
✅ Why deep learning was almost impossible until 2015 (Level 3)
✅ What manifolds are and why they matter (Level 4)
✅ The cutting-edge mHC technique that improves AI training (Level 5)

            Your Mission: Complete all 5 levels and you'll understand a paper that most computer science 
            PhD students find challenging. No joke—this is university-level material made accessible.
          

Contemporary Applications and Future Directions

Machine learning has become ubiquitous in modern computing systems. The following table illustrates representative applications across various domains:

Domain	Application	ML Technique
Computer Vision	Facial recognition, autonomous driving	Convolutional Neural Networks
Natural Language	Machine translation, text generation	Transformer architectures
Recommendation	Content curation, personalized advertising	Collaborative filtering, deep learning
Game Playing	Chess, Go, video game AI	Reinforcement learning
Scientific Discovery	Drug discovery, protein folding	Graph neural networks

Course Trajectory

This course provides a structured progression from foundational concepts to contemporary research:

Level 1: Pattern recognition and supervised learning fundamentals
Level 2: Neural network architectures and activation functions
Level 3: Residual connections and optimization challenges
Level 4: Differential geometry and manifold theory
Level 5: Analysis of mHC: Manifold-Constrained Hyper-Connections

The objective is to develop sufficient mathematical maturity and domain knowledge to comprehend state-of-the-art research in deep learning architecture design.

Try It Yourself!

🧠 Quick Check: Pattern Recognition

Look at the sequence below. What comes next?

2, 4, 6, 8, 10, ?

Answer: 12! You recognized the pattern: add 2 each time.

This is exactly what AI does—find the rule that connects inputs to outputs.

Challenge Question

A computer is learning to predict temperatures. It sees:
Monday: 20°C → Tuesday: 22°C
Tuesday: 22°C → Wednesday: 24°C
Wednesday: 24°C → Thursday: 26°C

What will the AI predict for Friday if Thursday was 26°C?
(Hint: Look for the pattern!)

Exercise: Inductive Reasoning

Given the arithmetic sequence defined by the recurrence relation:

              an = an-1 + 2, with a1 = 2
            

Determine the value of a₆.

Solution: By direct computation, the sequence is 2, 4, 6, 8, 10, 12. Therefore, a₆ = 12.

This exemplifies the inductive inference task central to supervised learning.

Problem: Linear Prediction

Consider a temperature prediction model with the following training data:
{(20, 22), (22, 24), (24, 26)} where the first value is day n and the second is day n+1.

Assuming a linear relationship T_n+1 = w·T_n + b, determine the values of w and b, then predict T_Friday given T_Thursday = 26.