rohitg00--ai-engineering-from-scratch
wehub-resource-sync
79 行
3.2 KiB
JSON
79 行
3.2 KiB
JSON
{
|
|
"lesson": "21-fairness-criteria-group-individual-counterfactual",
|
|
"title": "Fairness Criteria - Group, Individual, Counterfactual",
|
|
"questions": [
|
|
{
|
|
"stage": "pre",
|
|
"question": "Which best matches the demographic-parity definition for groups A=a and A=a'?",
|
|
"options": [
|
|
"Equal true-positive and false-positive rates across groups",
|
|
"Equal predictive value across groups",
|
|
"Equal Lipschitz constant across groups",
|
|
"Equal acceptance rates: P(Y=1 | A=a) = P(Y=1 | A=a')"
|
|
],
|
|
"correct": 3,
|
|
"explanation": ""
|
|
},
|
|
{
|
|
"stage": "check",
|
|
"question": "What does the Chouldechova / Kleinberg-Mullainathan-Raghavan 2017 impossibility result say?",
|
|
"options": [
|
|
"Group fairness is always achievable by re-weighting",
|
|
"Under unequal base rates, demographic parity, equalized odds, and conditional use accuracy equality cannot all hold simultaneously",
|
|
"Counterfactual fairness implies demographic parity",
|
|
"Individual fairness implies counterfactual fairness for any DAG"
|
|
],
|
|
"correct": 1,
|
|
"explanation": ""
|
|
},
|
|
{
|
|
"stage": "check",
|
|
"question": "Dwork et al. 2012 define individual fairness via:",
|
|
"options": [
|
|
"Demographic parity at the individual level",
|
|
"Equal accuracy across protected groups",
|
|
"A Lipschitz condition on the decision map relative to a task-specific similarity metric, so similar individuals get similar decisions",
|
|
"A causal DAG with a sensitive-attribute intervention"
|
|
],
|
|
"correct": 2,
|
|
"explanation": ""
|
|
},
|
|
{
|
|
"stage": "check",
|
|
"question": "Counterfactual fairness (Kusner et al. 2017) requires:",
|
|
"options": [
|
|
"No causal assumptions",
|
|
"Equal Lipschitz constants for all groups",
|
|
"Only group-level statistics",
|
|
"A causal DAG; the decision is unchanged when the individual's sensitive attribute is counterfactually altered"
|
|
],
|
|
"correct": 3,
|
|
"explanation": ""
|
|
},
|
|
{
|
|
"stage": "post",
|
|
"question": "Why do backtracking counterfactuals (arXiv:2401.13935) matter for legal compliance?",
|
|
"options": [
|
|
"Instead of intervening on a protected attribute (which is legally problematic), they ask which combination of actual features would have produced the counterfactual outcome",
|
|
"They eliminate the need for any causal model",
|
|
"They replace embedding-based bias metrics",
|
|
"They prove the impossibility theorem is wrong"
|
|
],
|
|
"correct": 0,
|
|
"explanation": ""
|
|
},
|
|
{
|
|
"stage": "post",
|
|
"question": "What does the ICLR Blogposts 2024 philosophical reconciliation argue?",
|
|
"options": [
|
|
"The impossibility theorems are resolved by re-weighting",
|
|
"With an explicit causal graph, satisfying certain group-fairness measures entails counterfactual fairness, so the apparent opposition between families is partly an artifact of leaving the causal model implicit",
|
|
"Group and counterfactual fairness are unrelated",
|
|
"Only individual fairness is justifiable"
|
|
],
|
|
"correct": 1,
|
|
"explanation": ""
|
|
}
|
|
]
|
|
}
|