Parameters

  • Learned from data during training.
  • Updated by the algorithm.
  • Examples: weights, biases. Hyperparameters
  • Set before training.
  • Control how learning happens.
  • Examples: learning rate, batch size, epochs, regularization strength, number of layers.