引導式測驗 · 硬 等級
AI 專案的資料準備測驗
測試資料準備情況的五項檢查,包括選擇偏差、資料外洩、標籤協議和 GDPR 目的限制等法律限制。
問題 1 的 8
銀行僅對批准的申請人提供還款結果。這顯示了哪些資料準備問題?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 銀行僅對批准的申請人提供還款結果。這顯示了哪些資料準備問題?
- 重複記錄
- Selection bias that makes the data unrepresentative
- A slow API
- Label disagreement between annotators
- 模型訓練中的資料外洩是什麼?
- Data being stolen by attackers
- Storage costs rising over time
- Losing records during migration
- Features that contain information only available after the outcome being predicted
- GDPR 的哪一項原則限制將出於某一目的收集的資料重複用於不相容的新目的?
- Purpose limitation
- Right to data portability
- Accountability
- Storage limitation
- 在人工智慧專案的資料準備度中:Cohen 的 kappa 衡量什麼?
- The speed of a data pipeline
- The number of missing values
- Agreement between annotators, corrected for chance
- The cost of labeling per item
- 為什麼產生人工智慧助理的檢索需要尊重現有的存取權限?
- To make responses load faster
- So the assistant does not reveal documents a user is not allowed to see
- Because embeddings need permissions to be calculated
- To lower the number of documents stored
- Why should forecasting models usually be tested on data split by time rather than randomly?
- It mirrors real deployment and avoids training on future information
- 隨機分割是非法的
- Time splits need less data
- Random splits cannot be used with labels
- Which statement reflects a common misconception about data readiness?
- Readiness depends on the specific use case
- Quality only needs to be good enough for the task
- Legal rights to use data must be checked
- A large volume of data means an organization is ready for AI
- What does a 'datasheets for datasets' document record?
- The cost of each database server
- The model's accuracy on the test set
- A dataset's origin, composition, collection method and intended use
- The passwords for accessing the data