Demonstrate the Ability to Train and Inference CERTIFIED-AI-EXPERT Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A team has trained a sentiment analysis model and wants to deploy it as a REST API for a web application. Which of the following is a critical step in the deployment process?

   • AContainerize the model with its dependencies and expose an endpoint that accepts text input and returns the sentiment score.
   • BStore the model in a database and query it directly from the web application.
   • CConvert the model to a PDF document and share it with the development team.
   • DRetrain the model on the web application's server before serving requests.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A financial firm has deployed a fraud detection model. When a new transaction arrives, the model must output a fraud score. Which of the following steps is part of the inference process for this model?

   • ACompare the transaction to all historical transactions to find the most similar one and use its label.
   • BStore the transaction in the training dataset and re-run the entire training pipeline.
   • CRetrain the model on the new transaction before making a prediction.
   • DPreprocess the transaction data into the same format used during training, then feed it to the model and interpret the output score.
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A company has a trained NLP model that must serve inference requests with a strict latency requirement of under 100 ms. The model is large and runs on GPUs, but the budget is limited. Which of the following strategies best balances latency and cost?

   • AUse a serverless inference service that scales to zero when idle, but keep the model in its original size.
   • BUse model quantization and deploy on CPU-only instances with optimized inference libraries.
   • CDeploy the model on multiple GPUs in parallel to reduce latency per request.
   • DDeploy the full model on a single high-end GPU instance to guarantee the lowest latency.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   Which statement accurately differentiates training from inference in terms of computational requirements?

   • ATraining typically requires more compute and memory than inference
   • BTraining requires only CPU, while inference requires GPU
   • CInference typically requires more compute than training
   • DBoth require identical compute resources
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A company has a model that was trained on a large dataset. They now need to serve inference for a real-time application. The model's inference latency is too high. Which of the following is the most effective approach to reduce latency without retraining the model?

   • AIncrease the number of GPUs used for inference to parallelize each request.
   • BRetrain the model with a smaller dataset to make it faster.
   • CApply model compression techniques such as pruning and quantization.
   • DUse a larger batch size for each inference request.
   Reveal Solution Discussion Select an answer first