Responsible AI practicesFull notesSummaryRingkasanStoriesPracticeModel Trade-Offs9 exam-style questions on this lesson.Question 1 of 9An economist reads the coefficients of a linear regression model to understand how interest rates affect inflation. Which concept does this show?AExplainability through surrogate modelsBControllabilityCDifferential privacyDInterpretabilityCheck answerQuestion 2 of 9A news site's neural network is a black box. The team uses a model-agnostic method to discover it labels business articles about sports teams as "sports". Which concept is this?AExplainabilityBRobustnessCInterpretabilityDControllabilityCheck answerQuestion 3 of 9. Select two.Which TWO are model-agnostic methods used for explainability? (Select TWO.)ATemperatureBRegression coefficientsCPartial dependence plotsDSurrogate modelsEConfusion matricesCheck answer0 of 2 selectedQuestion 4 of 9A regulator requires that every decision be explained exactly from the model's internals. What is the likely trade-off?AUnlimited choice of deep neural networksBLimited choice of algorithms and usually lower performanceCHigher performance with no downside at allDNo need for any model documentationCheck answerQuestion 5 of 9A team needs high performance and only a general understanding of the model's behavior. What should it use?ANo model at allBAn interpretable linear model onlyCA higher-performing model with explainability methodsDA model trained on an air-gapped networkCheck answerQuestion 6 of 9A hospital adds differential privacy to its model to protect patient data. What is the trade-off?ATransparency improves, but safety dropsBBoth safety and transparency improve togetherCNeither changes, since privacy is separateDSafety improves, but the model becomes harder to inspectCheck answerQuestion 7 of 9Why are models trained on air-gapped networks harder to audit?ATheir isolation makes external audits harderBThey use different evaluation metricsCThey can't be deployed to productionDThey have more parameters than other modelsCheck answerQuestion 8 of 9What makes a model controllable?AIt has many hidden layers that learn featuresBChanging its training data changes its behavior as expectedCIt runs on a single server under one adminDIt ignores new training data once deployedCheck answerQuestion 9 of 9. Select two.Which TWO techniques improve controllability? (Select TWO.)ARemoving all monitoringBData augmentationCSwitching to a larger black box modelDTraining constraintsEEncrypting the model weightsCheck answer0 of 2 selectedTransparent and Explainable Models8 exam-style questions on this lesson.Principles of Human-Centered Design for Explainable AI8 exam-style questions on this lesson.