Q59 — AWS AIF-C01 Ch.2

Question 59 of 100 | ← Chapter 2

A data scientist is building a model to generate images of people in various occupations. The data scientist discovers that input data contains bias, and certain attributes influence image generation, introducing bias into the model.

• A. Data augmentation for imbalanced classes ✓
• B. Model monitoring for class distribution
• C. Retrieval-Augmented Generation (RAG)
• D. Image watermark detection

Correct Answer: A. Data augmentation for imbalanced classes

Explanation

When input data contains bias that propagates into generated images—especially across occupational categories—data augmentation for imbalanced classes (A) is an effective mitigation technique. By synthetically increasing representation of underrepresented groups or modifying samples to improve balance, this approach reduces the model’s reliance on biased patterns in the original data, thereby decreasing bias in outputs. Other options—such as class-distribution monitoring (B), RAG (C), or watermark detection (D)—do not directly address bias stemming from skewed training data distributions.