Practice question · True or false
Backpropagation is efficient because it reuses each layer's intermediate derivative instead of recomputing it for every weight.
Hints
- How many times would a naive chain-rule evaluation visit a layer?
- Shared subexpressions are computed once.
Show the answer
True
Why
True. The chain rule applied naively would recompute the same partial products once per weight; backpropagation sweeps backward once, caching each layer's error signal and reusing it for every weight feeding that layer. It is dynamic programming on the computation graph, which is what makes training large networks feasible.
Practise Differentiation Rules
The app has 6 more questions on this lesson, and keeps your place in the course. Computer Science I is free to start.