In the deep learning network, the backpropagation algorithm is used to find the optimal parameters. What rules are used in the backpropagation algorithm to obtain the derivation layer by layer?

Question

Options

• AChain rule
• BCumulative rule
• CLaw of reciprocity
• DRule of Normalization

• A
  78% (50)
• B
  6% (4)
• C
  3% (2)
• D
  13% (8)

Explanation

Option A (Chain rule) is correct because backpropagation computes gradients by working backwards through the network, multiplying partial derivatives layer by layer - which is exactly what the chain rule enables: differentiating a composition of functions by chaining together each function's local derivative.

Why the distractors are wrong:

• B (Cumulative rule) is not a standard calculus or ML concept; it conflates "accumulating" gradients with the actual mathematical rule that makes it possible.
• C (Law of reciprocity) refers to relationships between inverse operations or mutual dependencies (e.g., in physics or combinatorics), not to gradient computation.
• D (Rule of Normalization) relates to scaling values to a specific range or distribution (e.g., batch norm), which has nothing to do with computing derivatives.

Memory tip: Think "back-CHAIN-propagation" - the algorithm literally chains derivatives backward from the output loss to each weight, one layer at a time. If you remember that "chain" = chain rule, the answer is always A.

No community discussion yet for this question.