引導式測驗 · 中等 等級

醫療保健測驗中的聯邦學習

測試分散式訓練、原始資料局部性、隱私限制、站點異質性、外部評估和治理要求。

問題 1 的 8

What remains at participating sites in federated learning?

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。

  1. What remains at participating sites in federated learning?
    • No data or updates are transmitted
    • Only predictions are shared with patients
    • Raw data can remain local while model updates are shared
    • Every patient record moves to one central database
  2. Does federated learning automatically guarantee privacy?
    • No; updates and metadata may still leak information
    • Yes, because raw data never move
    • Yes, any model update is anonymous
    • No privacy controls can be used
  3. Why align feature and outcome definitions across hospitals?
    • To make every hospital use the same vendor
    • 消除驗證的需要
    • To ensure each site has identical patients
    • Different coding and measurement can create inconsistent training signals
  4. Which evaluation best tests generalization for a federated model?
    • Measuring communication speed alone
    • Testing at a site that did not contribute to training
    • Testing only on the largest training site
    • Reusing local training predictions
  5. How can large sites affect the federation?
    • They guarantee performance for small sites
    • They make privacy irrelevant
    • They may dominate updates if aggregation does not account for site differences
    • They automatically remove bias
  6. What can secure aggregation or differential privacy contribute?
    • A guarantee of identical site performance
    • Additional protections with possible utility trade-offs
    • A guarantee that the model is clinically valid
    • A replacement for access control
  7. What should be agreed before training begins?
    • Only the color of the dashboard
    • The final accuracy value
    • The model name alone
    • Task, data definitions, update protocol, security, and oversight responsibilities
  8. What does the EXAM study illustrate?
    • A multi-site federated training workflow for a specific COVID-19 task
    • A privacy guarantee for all medical AI
    • A universal model for every disease
    • A replacement for clinical evaluation