Hyperparameter
A setting chosen by engineers before training — not learned from data — that controls how learning runs.
Plain English Explanation
A hyperparameter is a setting chosen by the people building a model — not learned from data. It controls how training runs: how fast the system learns, how large each batch is, or how many layers the network has.
Tuning hyperparameters is often as important as choosing the right architecture.
Analogy
Hyperparameters are like the oven temperature and cooking time on a recipe — they are not ingredients, but they determine whether the dish turns out well or burns.
How is it used?
AI engineers tune hyperparameters when building everything from recommendation engines to large language models. Getting them wrong can mean a model that never learns properly or takes weeks instead of days.
Real-world Example
Choosing learning rate, batch size, and number of training epochs for a new chatbot are all hyperparameter decisions.
Common Misconceptions
Hyperparameters are not the same as model parameters (weights) — only the latter are updated automatically during training.