Neural Network

A neural network is a model made of simple units arranged in layers. Each unit multiplies its inputs by learned weights, adds them up, and passes the result on.

No single unit does anything clever. The capability comes from stacking thousands of them, so that early layers pick up simple patterns and later layers combine those into complicated ones.

Think of it like this. Think of a huge committee where every member votes on one narrow question and passes their verdict up a level. Nobody sees the whole problem. The decision emerges from how the votes are weighted.

How it works

• Inputs enter the first layer. Pixels, words, numbers, whatever the data is.
• Each connection carries a weight. A number saying how much that input matters to the next unit.
• Each unit sums and squashes. It adds up its weighted inputs, adds a bias, then applies an activation function that introduces a curve.
• Hidden layers stack in between. Each one works on the output of the last, building more abstract features.
• Training adjusts the weights. Predictions are compared to the truth, and every weight is nudged to reduce the error.

Common mistakes

• "It works like a brain." The inspiration is loose. An artificial neuron is arithmetic, not biology.
• "A neural network is deep learning." The network is the structure. Deep learning is the practice of using many layers and a lot of data.
• "More layers is better." Depth adds capacity and also adds overfitting, cost, and training difficulty.
• "The activation function is a detail." Without it the whole network collapses into one straight line, however many layers you stack.