Tracking and Running ML Experiments PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   A team is developing a custom ML model using TensorFlow and needs to run a series of experiments with different hyperparameters. They want to orchestrate the steps as a repeatable pipeline. Which Google Cloud environment is most appropriate for this task?

   • AVertex AI Workbench
   • BKubeflow Pipelines
   • CGemini Enterprise Agent Platform Pipelines
   • DExperiments on Gemini Enterprise Agent Platform
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   Why is managing model lineage important for governance?

   • AIt automatically improves model accuracy
   • BIt speeds up the deployment process
   • CIt helps reduce the cost of model training
   • DIt ensures that models can be audited and reproduced
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A machine learning engineer is using Experiments on Gemini Enterprise Agent Platform to track a series of fine-tuning runs for a large language model. They need to record the exact training dataset version, the base model checkpoint, and the hyperparameters for each run so they can later identify which configuration produced the best validation loss. What is the most efficient way to achieve this?

   • ACreate a separate experiment for each hyperparameter combination
   • BUse the experiment run's metadata parameters to log the dataset version, base model, and hyperparameters
   • CStore the information in a separate spreadsheet and manually link it to each experiment run
   • DOnly record the validation loss and rely on the timestamp to infer the configuration
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A regulated industry requires that every deployed model can be traced back to the exact training data, code version, and hyperparameters used. The team uses Gemini Enterprise Agent Platform Pipelines for training. What is the most effective way to ensure this lineage is recorded and accessible?

   • AUse a naming convention for model artifacts that includes the training data and code version
   • BManually document the lineage in a compliance spreadsheet after each deployment
   • CStore the pipeline run ID in the model artifact's metadata and use it to retrieve the full lineage
   • DOnly record the model artifact's URI and rely on the artifact store's timestamps
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A team is using LLM-as-a-judge to evaluate the quality of responses from two different chatbots. They notice that the judge consistently gives higher scores to responses from chatbot A, even when the responses are factually similar. They suspect the judge is biased toward the style of chatbot A. What is the most effective way to reduce this bias?

   • AProvide the judge with the chatbot names and ask it to be fair
   • BBlind the judge to the source of the responses by removing any identifying information
   • CUse a different LLM as the judge for each chatbot to balance the bias
   • DUse a single judge model and accept the bias as inherent
   Reveal Solution Discussion Select an answer first