問82
頻出度★★☆
ニューラルネットワークの学習に用いられるバックプロパゲーションで行われていることはどれか。
令和8年度 ITパスポート試験 問82
テクノロジ系基礎理論 › 情報に関する理論
This question's explanation and your answers so far.
解説
問 82
- ア
正解 Backpropagation (the error back-propagation method) is a learning technique that propagates the error between the neural network output and the correct answer backward from the output layer to the input layer, gradually adjusting the “weights” of the connections between nodes so that the error keeps getting smaller. Therefore option ア, “It adjusts the weight of each node to reduce the error”, is correct.
Incorrectイ
The part “it adjusts the weight of each node” is correct, but the purpose is wrong. The activation function is decided in advance during the network design stage; it is not something backpropagation selects during training. The purpose of backpropagation is solely to minimize the error.
Incorrectウ
What backpropagation adjusts is the “weights” between nodes, not the number of nodes. The number of nodes (the network structure) is decided by the designer before training; it is not increased or decreased during training to reduce the error.
Incorrectエ
Backpropagation neither changes the number of nodes nor aims to improve the processing speed. Both the adjustment target (the number of nodes) and the purpose (the processing speed) are wrong.
ここがポイント
- Backpropagation (the error back-propagation method) is a learning technique that propagates the error arising at the output side backward to the input side, updating the “weights” of each connection to minimize the error (loss).
- Distinguish clearly that the target of adjustment is the “weights”, not the number of nodes or the selection of the activation function.
用語