Configuration settings that control the learning process of machine learning algorithms but are not learned from data. Hyperparameters include learning rate, batch size, and network architecture choices, requiring careful tuning to optimize model performance and training efficiency.
