Picture a vast web of tiny switches, each lighting up a little more or a little less depending on what it receives from its neighbors. An artificial neural network is exactly that: a mathematical model, loosely inspired by the brain, that learns to turn inputs (an image, text, sound) into useful outputs (a number, a category, a translation).
The neuron, the building block
Each artificial neuron receives several signals, weights them, sums them, adds a bias, then passes the result through a non-linear activation function (such as ReLU or sigmoid). Mathematically:
$$y = f\left(\sum_{i=1}^{n} w_i x_i + b\right)$$
The weights $w_i$ are the parameters the network tunes during learning. It is the non-linearity of $f$ that lets the network approximate complex relationships.
Stacked layers
Neurons are organized into layers: an input layer, one or more hidden layers, and an output layer. When there are many layers, we speak of deep learning.
| Layer | Role |
|---|---|
| Input | Receives the raw data |
| Hidden | Extract increasingly abstract patterns |
| Output | Produces the final prediction |
How it learns
The network compares its output to the expected answer through a loss function, then corrects its weights via backpropagation and gradient descent, repeated over thousands of examples.
- More data and more layers richer patterns
- Risk of overfitting if the model memorizes instead of generalizing
A neural network is not programmed rule by rule: it learns its own rules from examples.