What is a neural network?
A neural network is a computing system that learns to recognise patterns by adjusting the strengths of connections between simple processing units. It is loosely inspired by how biological neurons work, but in practice it is a mathematical structure: layers of units (called neurons) connected by coefficients (called weights). When you give it an input—for example, the pixel values of an image—the network transforms that input through each layer, applying the weights and a simple mathematical function, until it produces an output, such as a label like “cat” or “dog”. The network does not start with any knowledge; it learns by being shown many examples. Each time it makes a prediction, it adjusts the weights (its parameters) to make the next prediction closer to the correct answer. That process of learning from data is called training.
How information flows through layers
A neural network is organised into layers: an input layer, one or more hidden layers, and an output layer. Each neuron in a layer receives signals from neurons in the previous layer, multiplies each signal by its own weight, sums them, and passes the result through a simple mathematical function—typically one that squashes the output into a fixed range, like 0 to 1. That output then becomes the input to the next layer. The pattern of connections and weights determines what the network pays attention to. Early layers detect simple features (like edges in an image), and deeper layers combine those into more complex patterns (like a nose or an ear). The output layer produces the final decision, such as a probability for each possible category.
How the network learns from data
The weights are the network’s parameters, and they are the only thing that changes during training. You start with random or small numbers for every weight. Then you feed the network a training example—a known input paired with the correct output. The network makes a guess, and a mathematical function measures how far off that guess is from the truth. That error is then used to adjust every weight, slightly, in the direction that would reduce the error on that example. Over thousands or millions of examples, these small adjustments accumulate, and the network’s accuracy improves. The key is that the network never memorises the examples; it learns general patterns that let it handle inputs it has never seen before.
Why it is called “neural”
The name comes from a loose analogy to the brain. A biological neuron receives electrical signals through its dendrites, sums them, and fires an output signal down its axon if the sum passes a threshold. A neural network’s artificial neuron does something similar: it sums weighted inputs and applies a threshold-like function. But the resemblance stops there. Real brains are vastly more complex—billions of neurons with trillions of connections, all interacting in ways we still do not fully understand. An artificial neural network is a simplified mathematical model, not a simulation of biology. The name stuck because the original researchers were inspired by neuroscience, but today’s networks are designed with engineering principles, not brain anatomy. Understanding them as layered mathematical functions, not as miniature brains, is the more useful way to think about them.