AI0-001 AI Concepts and Foundations Practice Question
A team deploying an AI model for real-time fraud detection notices that inference latency is too high. The model is a deep neural network with 50 layers, deployed on a cloud GPU. Which of the following is the BEST approach to reduce latency while maintaining acceptable accuracy?
⚠ Common exam trap
CompTIA often tests the misconception that simply upgrading hardware or reducing batch size is the best latency fix, when in fact architectural compression techniques like knowledge distillation are the most effective for deep models with strict latency budgets.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Apply knowledge distillation to create a smaller model.
Knowledge distillation trains a smaller 'student' model to mimic the behavior of a larger 'teacher' model, significantly reducing the number of parameters and layers while preserving most of the original accuracy. This directly addresses the high inference latency caused by the 50-layer DNN by producing a compact model that runs faster on the same GPU hardware.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Deploy the model on a more powerful GPU.
Why it's wrong here
A faster GPU raises throughput but the 50-layer network's serial depth still dominates per-inference latency; each layer must wait for the previous one. GPU upgrades suit throughput-bound batch workloads. Latency here needs model compression, pruning, quantisation or layer reduction, which cut the sequential computation itself.
- ✗
Reduce the batch size for inference.
Why it's wrong here
Real-time fraud scoring typically serves one transaction per request, so batch size is already 1 and cannot be reduced further. Tuning batch size is correct for throughput-oriented offline scoring, where larger batches amortise GPU kernel overhead. It does nothing for single-request latency on a 50-layer network.
- ✗
Replace the DNN with a logistic regression model.
Why it's wrong here
Swapping the deep network for logistic regression discards the non-linear feature interactions that detect fraud, breaking the accuracy requirement. Logistic regression suits simple, linearly separable scoring with strict interpretability needs. The stem demands latency reduction while maintaining acceptable accuracy, which distillation or pruning preserves.
- ✓
Apply knowledge distillation to create a smaller model.
Why this is correct
Knowledge distillation trains a smaller student network to mimic the 50-layer teacher, cutting inference computation and latency while retaining most accuracy. This satisfies the stem's constraint by reducing depth and parameters rather than merely quantising or batching the existing model.
About these practice questions
One of 962 original AI0-001 practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI0-001 practice question is part of Courseiva's free CompTIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI0-001 exam.