In the process of training the neural network, we use the gradient descent method to continuously update which value, which makes the loss Function minimization?

Question

Options

• ANumber of samples
• BEigenvalues
• CHyperparameter
• Dparameter

• A
  1% (1)
• B
  6% (4)
• C
  9% (6)
• D
  84% (56)

Explanation

Parameters (weights and biases) are the values gradient descent updates during training - these are the internal learnable values of the model that get adjusted iteratively to reduce the loss function toward a minimum.

Why the distractors are wrong:

• A (Number of samples): This is fixed by your dataset; changing it would mean collecting more data, not training.
• B (Eigenvalues): These are mathematical properties of matrices, not values the training loop updates.
• C (Hyperparameters): These (like learning rate, batch size, number of layers) are set before training by the developer and are not touched by gradient descent.

Memory tip: Think of parameters as the model's adjustable knobs that the algorithm turns during training, while hyperparameters are the settings on the machine itself that you configure before you start - gradient descent only touches the knobs, never the machine settings.

No community discussion yet for this question.