Identifying Risks to AI Solutions PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   Which metric is commonly used to monitor bias by comparing prediction rates across different demographic groups?

   • ALatency
   • BThroughput
   • CDemographic parity
   • DMean squared error
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A credit union deploys a tabular model on Vertex AI Agent Platform for loan decisions. A regulator asks for an explanation of why a specific applicant was rejected. The data science team needs to generate a feature-attribution explanation for this individual prediction. Which approach should they use?

   • ACall the Agent Platform Inference `explain` method on the deployed model to get feature attributions for the specific instance.
   • BRetrain the model using a simpler, interpretable algorithm like logistic regression to make the decision process transparent.
   • CQuery the training dataset to find the most similar historical applicant and present their outcome as the explanation.
   • DUse the Vertex AI Model Garden to find a pre-trained explainable model and compare its output with the deployed model.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A marketing team uses a public LLM to draft campaign emails. The team lead is concerned that an employee might accidentally paste a list of customer email addresses into the prompt. The company wants a technical control that works at the endpoint level to block this specific type of data from being sent. Which control is the most appropriate?

   • ADeploy a Cloud DLP job that scans the company's Cloud Storage buckets for email addresses and flags them.
   • BUse a DLP-integrated web proxy that inspects outbound HTTPS requests and redacts or blocks content matching email address patterns.
   • CConfigure a Cloud NAT gateway with a firewall rule that blocks traffic to the LLM's IP address.
   • DRequire employees to use a company-approved VPN that logs all traffic to the LLM.
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   Which tool is specifically designed to protect AI systems by filtering both input and output content for safety and security risks?

   • ACloud Monitoring
   • BModel Armor
   • CCloud Load Balancing
   • DCloud CDN
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A financial services firm allows its analysts to use a public LLM chat tool for research. The compliance team is worried that an analyst might paste a customer's personally identifiable information (PII) into the chat. Which control is most effective at preventing this specific accidental data leak?

   • ARequire analysts to sign an acceptable use policy that prohibits sharing customer data with external tools.
   • BConfigure a web proxy with a DLP integration that inspects outbound HTTPS requests and blocks any content matching PII patterns.
   • CDeploy a Cloud DLP API inspection job that scans BigQuery tables for PII and flags them for the data governance team.
   • DEnable VPC Service Controls on the project that hosts the LLM API endpoint.
   Reveal Solution Discussion Select an answer first