In neural networks, which of the following methods are used to update the parameters when training the network to minimize the loss function?

Question

Options

• AForward propagation algorithm
• BPooling calculation
• CConvolution calculation
• DBackpropagation algorithm

• A
  11% (6)
• B
  2% (1)
• C
  6% (3)
• D
  81% (43)

Explanation

Backpropagation (D) is the algorithm that updates network parameters during training. It works by computing the gradient of the loss function with respect to each weight using the chain rule, then adjusting weights in the direction that reduces loss - typically via gradient descent or an optimizer like Adam.

Why the distractors are wrong:

• A (Forward propagation) passes inputs through the network to generate predictions, but makes no weight updates - it's the "read" pass, not the "learn" pass.
• B (Pooling) is a spatial downsampling operation in CNNs (e.g., max pooling) that reduces feature map size - it has no role in parameter updates.
• C (Convolution) is a feature extraction operation that applies learned filters to inputs - again, it's part of the forward pass architecture, not the training update mechanism.

Memory tip: Think "Back = update." Forward prop goes forward to make a prediction; backprop goes backward to fix mistakes. Only the backward pass touches the weights.

No community discussion yet for this question.