安全与可信 AI
Your differentially private model lost significant accuracy. How do you balance privacy and utility?
原题One malicious participant is poisoning your federated learning model. How do you defend against it?
原题Your AI hiring model uses proxy features for protected attributes. How do you eliminate proxy discrimination?
原题Your predictive model creates a feedback loop of biased outcomes. How do you break it?
原题Your AI generates fake news images. How do you implement watermarking for AI-generated content?
原题Your AI denies a service, and the user has no way to challenge it. How do you design an appeals process?
原题An auditor asks why your AI rejected a request 6 months ago, and you have no logs. How do you build audit trails?
原题You removed PII, but users were re-identified from anonymized data. How do you prevent re-identification?
原题A pre-trained model from an open-source repo may contain a hidden backdoor. How do you detect it?
原题Your LLM's training data was deliberately poisoned by an adversary. How do you respond?
原题
暂无题目