Quiz collegato alla guida · Medio Livello

Kit di strumenti per l'equità: AIF360, Fairlearn e Aequitas Quiz

Confronta AIF360, Fairlearn e Aequitas in base ai loro ruoli di valutazione e mitigazione, input, limiti e relazione con la revisione legale.

Domanda 1 di 8

Which description best fits AI Fairness 360?

Anteprima di ogni domanda8 domande

Le risposte e le spiegazioni vengono visualizzate mentre rispondi al quiz.

  1. Which description best fits AI Fairness 360?
    • A law-enforcement database
    • A tool that guarantees legal compliance
    • A toolkit with fairness metrics and pre-, in-, and post-processing algorithms
    • A single metric for hiring
  2. What does Fairlearn MetricFrame help users do?
    • Disaggregate metrics across groups and compare performance
    • Generate model weights from scratch
    • File a legal complaint
    • Remove all sensitive data
  3. Which role does Fairlearn ExponentiatedGradient serve?
    • Post-hoc score thresholding
    • Generating legal notices
    • Replacing evaluation data
    • Training under a selected fairness constraint and objective
  4. Which task is Aequitas designed to support?
    • Training neural networks
    • Auditing bias with group disparity metrics and reports
    • Selecting data licenses
    • Explaining every individual prediction
  5. What must be correctly specified before using toolkit metrics?
    • Protected/group attributes, labels, favorable outcomes, and decision threshold
    • Only the package name
    • The model’s user interface
    • The operating system wallpaper
  6. How does Fairlearn ThresholdOptimizer adjust a fitted model?
    • It removes the model’s validation set
    • It guarantees equal accuracy for every group
    • It applies group-specific thresholds under a chosen constraint and objective
    • It replaces a model with a legal rule
  7. Does a fairness toolkit certify legal compliance?
    • Yes, by default
    • No; toolkits provide measurements or methods, not legal approval
    • Only when using open-source models
    • Yes, if two libraries agree
  8. What should follow an evaluation that identifies a disparity?
    • Treat the metric as legal approval
    • Ignore the result if overall accuracy is high
    • Automatically delete the model
    • Interpret the context, assess tradeoffs, and decide on a justified mitigation or use limit