Hyperparameter
A setting you choose before training (learning rate, batch size, layers) — as opposed to the weights the model learns.
A hyperparameter is a configuration value you set before training that controls the learning process itself — learning rate, batch size, number of layers, regularization strength — as opposed to the model’s parameters (weights), which are learned from data. Hyperparameters shape how, and how well, the parameters get learned.
Worked example: the learning rate is the highest-leverage hyperparameter — too high and training diverges, too low and it crawls; finding good values is hyperparameter tuning, done by grid search, random search, or Bayesian optimization. Gotcha: hyperparameter search is expensive (each trial is a full training run) and easy to overfit to the validation set if you tune too aggressively — which is why you hold out a final test set the tuning never touches, and why sensible defaults plus tuning only the few high-impact knobs beats brute-forcing everything.