Google Professional Machine Learning Engineer Question 206
Select 3Google Cloud PlatformYou are tasked with evaluating a generative AI solution that produces synthetic images for an e-commerce platform. The goal is to ensure the generated images meet business requirements such as high visual fidelity, product consistency, and diversity. Which metrics or evaluation techniques would be most appropriate to assess the model's performance?
- A
Fréchet Inception Distance (FID)
- B
BLEU score
- C
Human qualitative evaluation through user surveys
- D
Diversity score for generated images
- E
Confusion matrix
Show answer and explanation
Correct answers: A, C, D
Explanation
Evaluating generative AI solutions requires specific metrics that align with the type of output being generated. For synthetic images, Fréchet Inception Distance (FID) assesses visual fidelity and realism, while diversity score helps ensure a broad range of outputs. Additionally, human qualitative evaluation provides subjective but essential feedback on the generated images' alignment with business needs. Metrics like BLEU or confusion matrix are irrelevant since they pertain to text-based models or classification tasks, respectively.
- A. Correct.
Fréchet Inception Distance (FID) is a widely used metric to evaluate the quality of generated images by comparing their distribution with real images. It is relevant to assessing visual fidelity and realism.
- B. Incorrect.
BLEU score is a metric used to evaluate text-based generative models, such as machine translation or text summarization, and is not applicable to image generation tasks.
- C. Correct.
Human qualitative evaluation through user surveys provides insights into how well the generated images align with business and user expectations, making it an important evaluation technique.
- D. Correct.
Diversity score is crucial for assessing whether the generative model produces a wide range of unique and varied images, which is important for avoiding repetitive outputs in an e-commerce context.
- E. Incorrect.
Confusion matrix is used for classification tasks and does not apply to evaluating generative AI models producing images.