Book 08 · AI & Machine LearningPage 14 of 52

Deep Learning

In Turkish: Derin Öğrenme

Updated September 30, 20262 min read

In short

Deep learning is a subset of machine learning that uses neural networks with many layers to learn complex patterns from raw data such as images and text.

What is deep learning?

Deep learning is a branch of machine learning based on neural networks with many hidden layers, which is where the word deep comes from. Each layer transforms its input into a slightly more abstract representation, so the network can learn complex patterns step by step. Modern deep learning models range from millions to hundreds of billions of parameters.

A key advantage of deep learning is that it discovers useful features on its own. In classic machine learning, engineers often had to hand-design features, such as counting edges in an image; a deep network given raw pixels learns to detect edges in its early layers, shapes in the middle layers, and whole objects like faces or cars in later layers. Training these large models requires a lot of data and specialized hardware such as GPUs, which can perform many calculations in parallel.

An analogy is learning to read: first you recognize strokes, then letters, then words, and finally the meaning of whole sentences, with each level building on the one before. Deep learning powers image and speech recognition, machine translation, the perception systems of self-driving cars, recommendations, and large language models.

Deep learning is often used interchangeably with AI or machine learning, but it is a subset of both: AI is the broad goal, machine learning is learning from data, and deep learning is machine learning with deep neural networks. For small, structured datasets, such as a spreadsheet of customer records, simpler machine learning methods are often faster, cheaper, and just as accurate.

Key takeaways

Deep learning uses neural networks with many hidden layers.
Each layer learns a more abstract representation of the data.
It learns features automatically from raw data such as pixels, audio, or text.
It needs large datasets and parallel hardware such as GPUs.
For small tabular datasets, classic machine learning is often a better fit.

Example

A deep network as stacked layers in NumPypython

import numpy as np

rng = np.random.default_rng(0)
layer_sizes = [784, 256, 128, 64, 10]  # input, three hidden layers, output

# One weight matrix per pair of neighboring layers (random until trained)
weights = [rng.normal(0, 0.1, (a, b)) for a, b in zip(layer_sizes, layer_sizes[1:])]

def forward(x):
    for w in weights[:-1]:
        x = np.maximum(0, x @ w)  # ReLU activation in each hidden layer
    return x @ weights[-1]        # output layer: one score per class

image = rng.random(784)           # a fake 28x28 image, flattened
print(forward(image).shape)       # (10,) -> scores for the digits 0-9

Readers ask

What is the difference between deep learning and machine learning?

Deep learning is a subset of machine learning that uses neural networks with many layers. Classic machine learning often relies on hand-picked features and works well on smaller, structured data, while deep learning learns features itself and excels at images, audio, and text.

Why is it called deep learning?

The word deep refers to the number of layers in the neural network. A network with many hidden layers between its input and output is considered deep, as opposed to a shallow network with only one or two.

Why does deep learning need GPUs?

Training a deep network involves huge numbers of matrix multiplications, and GPUs can perform thousands of these calculations in parallel. This makes training many times faster than on a regular CPU.

Often compared

Machine Learning vs Deep LearningMachine learning covers algorithms that learn from data; deep learning is a subset whose many-layered neural networks find features alone but need more data.