IPA IP 2026 Question 82
A question IPA Japan published in 令和8年度 ITパスポート試験 公開問題 問82, translated into English. ITPEC is based on the same exams.
Which of the following is performed by backpropagation, which is used for training a neural network?
Explanation
Adjusting the weights to reduce the error — backpropagation is how a neural network learns from the difference between its output and the correct answer.
Training repeats two passes. In the forward pass the input flows through the layers and the network produces an output. The error between that output and the correct answer is then sent backwards, from the output layer towards the input layer, and each weight is changed a little in the direction that makes the error smaller (gradient descent). Repeated over many training examples, the weights settle at values that give small errors.
Answer (a)
Why not others:
- •(b) — the activation function (sigmoid, ReLU and so on) is chosen by the designer when the network is built; backpropagation uses it but does not select it
- •(c) — the number of nodes is part of the network's structure, fixed before training; backpropagation changes the weights, not the structure
- •(d) — changing the structure for speed is a design decision, and the aim of backpropagation is accuracy, not processing speed
Key rule: In a neural network the structure (layers, nodes, activation function) is designed; the weights are learned, and backpropagation learns them by propagating the output error backwards to reduce it.
Basic Theory · AI-written, may contain errors
This question comes from the past questions IPA Japan publishes, translated into English by ITPEC Practice. ITPEC Practice is an independent study tool and is not affiliated with IPA or ITPEC. See the original paper (PDF) or Report an issue.