引导式测验 · 中等 等级

AI 异常检测测验

查看异常分数、无监督学习、情境和集体案例、阈值和工业监控限制。

相关引导路径人工智能异常检测
问题 1 的 8

为什么异常检测经常在没有标记异常示例的情况下发挥作用？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 为什么异常检测经常在没有标记异常示例的情况下发挥作用？
    • 异常现象很少见且不可预测，因此模型会学习“正常”
    • 标签的使用是非法的
    • 异常现象永远不会发生
    • 它仅使用图像
  2. “上下文异常”是指以下数据点：
    • 处处总是不正常
    • 在一种情况下正常，但在另一种情况下则异常
    • 数据库中的拼写错误
    • 重复记录
  3. What does the cited WACV autoencoder study use as an anomaly score?
    • A legal fraud label assigned to every record
    • A guarantee that no future fault will occur
    • Reconstruction error from an autoencoder trained to model the data
    • The number of random trees in an Isolation Forest
  4. How does the original Isolation Forest paper isolate candidate anomalies?
    • It requires an expert to label every anomaly in advance
    • It measures reconstruction error from a neural decoder
    • It uses a fixed threshold that proves an item is defective
    • Random partitioning tends to isolate anomalous instances in fewer splits
  5. 过于敏感地设置异常阈值的主要危险是什么？
    • 分析师充斥着虚假警报
    • 模型运行速度更快
    • 数据被加密
    • 异常现象变得不可能
  6. What can an unusual machine-vibration score reasonably trigger?
    • An automatic shutdown because failure is guaranteed
    • A maintenance investigation or additional measurement, without predicting failure with certainty
    • A confirmed diagnosis of the failed component
    • A claim that the machine will fail within a fixed number of hours
  7. Why is an anomaly score not automatically a fraud or defect label?
    • It is always a verified label
    • Anomaly scores cannot be compared
    • It indicates unusualness under a model or baseline, not the cause or consequence
    • The threshold proves intent
  8. What should be documented when setting an anomaly threshold?
    • 仅模型的名称
    • A claim that false positives are impossible
    • The number of GPUs alone
    • The reference data, score definition, threshold rationale, and review process