What Is a Neural Network?
Learn how a neural network turns numerical inputs into predictions and improves through feedback.
What Is a Neural Network? explained visually
Follow a dog-and-zebra example to see how pixels, layers, weights, and feedback produce a prediction.
What you will learn
Build an intuitive understanding of neurons, layers, learning, and common machine-learning tasks.
Learn from examples
A neural network learns patterns from labeled examples. During training, it compares each prediction with the correct answer and uses the difference as feedback.
Build features through layers
An image enters the network as numerical pixel values. Early layers respond to simple patterns such as edges and brightness changes, while deeper layers combine those patterns into more useful features.
Control influence with weights
Connections between artificial neurons contain weights. A weight controls how strongly one signal influences the next calculation, allowing the network to emphasize useful evidence.
Improve through feedback
When a prediction is wrong, training measures the error and adjusts the network's parameters. Repeating this process across varied examples can improve performance on new data.
Quick check: test your understanding
What does an image network receive first?
Numerical pixel values rather than complete concepts such as dog or zebra.
What does a weight control?
How strongly one signal influences another part of the network.
How does the network improve?
Training compares predictions with correct answers and adjusts parameters to reduce error.
Accessible referenceLesson 1 transcript355 words · Matches the published narration and captionsOpen transcript
Imagine you want to teach a computer the difference between a dog and a zebra. You show it many pictures and provide the correct answer each time. In the beginning, the computer may guess. After enough examples and corrections, its answers begin to improve.
The system doing this work can be a neural network. A neural network is a computer model that learns patterns from examples. It contains many small processing units called artificial neurons. These neurons pass numerical information to one another.
When the network receives a picture, it does not immediately see a dog or a zebra. It starts with pixels. Each pixel becomes a number that describes information such as brightness and color.
The first layer looks for very simple clues. Some neurons may react to edges. Other neurons may react to light and dark areas. The next layer combines those clues into larger shapes. It may notice curves, ears, or stripes. Deeper layers combine those features and begin to recognize the complete animal.
The connections between neurons contain values called weights. Think of each weight as an importance control. A strong weight tells the network to pay more attention to a signal. Zebra stripes may receive strong importance when the network tries to recognize a zebra.
Now suppose the network sees a zebra but predicts a dog. During training, we compare its prediction with the correct answer. The network measures how wrong it was and adjusts its weights slightly. Then it tries again. Repeating this process with many examples helps the network make better predictions.
Neural networks take some inspiration from biological neurons, but they are far simpler than a human brain. They are mathematical systems built to recognize patterns.
You already encounter neural networks in everyday life. They can help a phone recognize a face, detect spam in email, understand spoken words, or find cars and people in a street image.
The important idea is simple. A neural network receives numbers, builds useful patterns through layers, makes a prediction, and improves through feedback. In the next lesson, we will look inside one artificial neuron and explain inputs, weights, and bias.