AI Deep Learning (DL)

Introduction of Deep Learning (DL)

The advancement of deep learning emerged around 2010.

Since then, it has tackled numerous challenges once deemed impossible.

This breakthrough did not originate from a singular event but from multiple converging advancements:

• Computational power reached new heights
• Storage capabilities significantly increased
• Enhanced training techniques surfaced
• Improved optimization strategies evolved

Neurons

Researchers estimate the human brain comprises between 80 to 100 billion neurons.

These neurons form an intricate web of hundreds of billions of interconnections.

Neurons, also known as nerve cells, serve as the foundation of the brain and nervous system.

They process input from external stimuli, transmit commands to muscles, and relay electrical impulses internally.

Neural Networks

Artificial Neural Networks, commonly referred to as Neural Networks (NN), are multi-layered perceptrons.

The perceptron concept laid the groundwork for modern deep learning architectures.

Neural Networks represent a revolutionary breakthrough in computational intelligence.

These models can solve complex challenges that traditional algorithms cannot, including:

• Disease diagnosis
• Facial recognition
• Speech interpretation

Neural Network Structure

Input data (yellow) is processed through a hidden layer (blue), refined further in another hidden layer (green), and ultimately produces an output (red).

Tom Mitchell

Tom Michael Mitchell, an American computer scientist, serves as a University Professor at Carnegie Mellon University (CMU).

Previously, he led CMU’s Machine Learning Department.

He defined learning as:

"A program learns from experience (E) concerning a task (T) and a performance measure (P) if its performance (P) improves with experience (E) on task (T)."

Example:

E: Repeated practice
T: Operating a vehicle
P: Efficiency of driving

Giraffe Chess Engine

In 2015, Matthew Lai, a student at Imperial College London, created Giraffe, a neural network-based chess engine.

Within 72 hours, Giraffe attained an international master’s skill level.

Unlike traditional chess engines requiring years of refinement, Giraffe's deep learning approach enabled rapid learning and adaptation.

Deep Learning vs. Traditional Computing

Classical programming follows:

Data + Algorithm = Output

Machine learning inverts the approach:

Data + Output = Algorithm

Machine Learning

Often mistaken for Artificial Intelligence, Machine Learning is actually a subset of AI.

It leverages data to enable systems to self-improve.

Arthur Samuel (1959) described it as:

"A field of study that enables computers to learn without explicit programming."

Intelligent Decision Process

• Record prior actions and results
• Simulate potential outcomes
• Compare past and present actions
• Assess whether the new action is beneficial
• Select the best available option
• Repeat continuously

Computers' ability to iterate this process millions of times has proven their capacity for advanced decision-making.