NCA-GENM Question 79
Single answerYou are tasked with evaluating the performance of two generative AI multimodal models designed for text-to-image generation. Model A uses a transformer-based architecture, while Model B employs a diffusion-based architecture. During evaluation, you measure metrics such as Fréchet Inception Distance (FID) and user feedback. Model A has a lower FID score compared to Model B, but Model B receives higher user satisfaction ratings. How should you interpret these results to choose the best model?
- A
Select Model A because FID is an objective metric and indicates better quality.
- B
Select Model B because user satisfaction is more important than FID in real-world applications.
- C
Combine both metrics to make a decision, as each provides different insights into model performance.
- D
Ignore both metrics and retrain both models using a different architecture.
Show answer and explanation
Correct answer: C
Explanation
When evaluating generative AI models, it's essential to consider both objective metrics, like FID, which evaluate the technical performance of the model, and subjective metrics, like user feedback, which reflect real-world usability and satisfaction. By combining these metrics, you can make an informed decision that balances technical quality and user needs.
- A. Incorrect.
FID is an objective metric that measures similarity between generated images and real images, but it alone may not fully capture user preferences or real-world usability.
- B. Incorrect.
User satisfaction is important, but relying solely on subjective feedback can overlook objective quality issues identified by metrics like FID.
- C. Correct.
Combining both metrics allows for a more comprehensive evaluation by considering both objective performance and real-world user preferences.
- D. Incorrect.
Ignoring both metrics and retraining the models would be inefficient and unnecessary unless the models consistently fail to meet evaluation criteria.