Generative AI Leader Google Cloud's Generative AI Offerings Practice Question
A healthcare startup is using Vertex AI Imagen to generate synthetic medical images for training a diagnostic model. The images must comply with HIPAA regulations and cannot contain any real patient data. The team fine-tuned Imagen on a dataset of de-identified medical scans. However, during testing, they notice that some generated images closely resemble specific patients from the original dataset, even though the dataset was de-identified. They suspect that the model memorized some training examples. The team needs to address this issue without losing image quality. They have access to the original training data and Vertex AI tools. What action should they take?
⚠ Common exam trap
Google often tests the misconception that post-processing or filtering can solve memorization, when in fact the root cause is in the training algorithm itself, requiring a privacy-preserving technique like differential privacy.
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
✓
Re-tune the model using differential privacy (DP-SGD) to prevent memorization of individual examples.
Differential privacy (DP-SGD) during fine-tuning adds calibrated noise to the gradient updates, which mathematically bounds the model's ability to memorize any single training example. This directly addresses the memorization issue while preserving the utility of the generated images, as the noise is carefully controlled to maintain overall image quality. Vertex AI supports DP-SGD through its custom training infrastructure, making it a practical choice for HIPAA-compliant medical imaging.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Use a post-processing step to blur or distort generated images.
Why it's wrong here
Blurring or distorting outputs destroys the diagnostic detail the training images require, defeating the model's purpose. It is tempting because post-processing is a familiar privacy technique, and it would be correct for de-identifying non-clinical images where fine anatomical detail is irrelevant.
- ✓
Re-tune the model using differential privacy (DP-SGD) to prevent memorization of individual examples.
Why this is correct
DP-SGD adds calibrated noise during fine-tuning, bounding any single training example's influence so the model cannot memorise identifiable scans. This removes the resemblance to specific patients while preserving overall image quality, meeting the HIPAA constraint without discarding the de-identified dataset.
- ✗
Increase the size of the training dataset by adding more synthetic images.
Why it's wrong here
Adding synthetic images does not remove memorised training examples; the model still reproduces them, so HIPAA risk persists. Augmentation suits class imbalance or dataset scarcity, not privacy remediation. The scenario needs deduplication or differential privacy applied to the original training data.
- ✗
Apply stricter output safety filters to block images that look like any known patient.
Why it's wrong here
Output filters block likenesses after generation but cannot remove the memorised training data causing them, so the privacy breach persists. It is tempting because safety filters are the standard Vertex AI output control, and they would be correct for blocking disallowed content categories rather than training-data memorisation.
Go deeper
Related to this question
About these practice questions
Courseiva writes every Generative AI Leader question from scratch — 1,008 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This Generative AI Leader practice question is part of Courseiva's free Google Cloud 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 Generative AI Leader exam.