問8
頻出度★★☆
AIの機械学習で利用するデータの取扱いに関する記述のうち, バイアスの低減やデータの品質を確保するために考えられる対策として, 適切なものだけを全て挙げたものはどれか。 a 学習の目的に適したデータであることを確認する。 b データの入手元・作成来歴を確認する。 c データへのアノテーションの付与は学習目的に合わせて実施する。 d 人間の目でも同定が困難と考えられる画像認識用のデータは除外する。
令和7年度 ITパスポート試験 問8
ストラテジ系ビジネスインダストリ › ビジネスシステム
This question's explanation and your answers so far.
解説
問 8
Incorrectア
ア (a, b) lacks c and d. Adding annotations in accordance with the learning objective (c) and excluding ambiguous data that is hard to identify (d) are both effective measures for reducing bias and ensuring quality; because these are omitted, it cannot be said to list all and only the appropriate items.
- イ
正解 The correct answer is イ (a, b, c, d). To reduce bias and ensure data quality in AI machine learning, all four statements are effective measures. a: Confirming whether the data is suitable for the learning objective prevents biased data that does not match the objective from being mixed in. b: Confirming the source of acquisition and creation history (provenance) lets you exclude untrustworthy or biased data sources, thereby ensuring quality. c: Adding annotations (labeling) in accordance with the learning objective prevents learning bias caused by incorrect or inconsistent labels. d: Image data that is difficult to identify even by the human eye is ambiguous and noisy, producing mislabeling and inconsistent judgments, so excluding it improves data quality and suppresses bias. Since a through d are all appropriate, イ, which lists all of them, is the correct answer.
Incorrectウ
ウ (a, d) lacks b and c. Confirming the source of acquisition and creation history (b) is a measure needed to ensure the trustworthiness of the data source, and annotation aligned with the objective (c) is a measure needed to ensure label quality; because these are missing, it is inappropriate.
Incorrectエ
エ (b, c, d) lacks a. Confirming whether the data is suitable for the learning objective (a) is the most basic measure for preventing biased data that does not match the objective from being mixed in; because it is not included, it cannot be said to list all of them.
ここがポイント
- Basic measures for ensuring data quality and reducing bias in AI machine learning: confirming suitability for the objective, confirming the source of acquisition / creation history (provenance), adding annotations aligned with the objective, and excluding ambiguous data that is hard to identify are all effective, so a through d are all correct.
- In questions of the “lists all and only the appropriate items” type, verify each element in each choice one by one, and select the combination that contains no incorrect element and omits no correct element.
用語