NCA-GENM Question 57
Select 3You are working on a generative AI multimodal project that involves text and image data. During the data preparation phase, you discover that the image dataset contains inconsistent resolutions, missing metadata, and some corrupted files. Meanwhile, the text dataset has missing values and inconsistent formatting. Which steps should you take to prepare the datasets for modeling?
- A
Normalize the image resolutions and remove corrupted image files.
- B
Impute missing values in the text dataset and standardize text formatting.
- C
Train the model directly on the raw datasets, as generative AI models typically handle noisy data.
- D
Use data augmentation techniques to increase the diversity of the image dataset.
- E
Generate synthetic metadata for the image dataset to fill in missing values.
Show answer and explanation
Correct answers: A, B, D
Explanation
Data preparation is a critical step in generative AI workflows. For multimodal projects involving both text and image data, ensuring consistency, handling missing values, and improving data diversity are essential. Normalizing resolutions, imputing missing text values, and applying data augmentation are standard practices to prepare datasets for effective modeling. Skipping data preparation or introducing synthetic metadata can lead to suboptimal or unreliable model performance.
- A. Correct.
Correct: Normalizing image resolutions and removing corrupted files are crucial steps in ensuring that the image data is consistent and usable for model training.
- B. Correct.
Correct: Handling missing values and standardizing text formatting are necessary steps to clean and prepare the text dataset for effective modeling.
- C. Incorrect.
Incorrect: Training on raw, unprocessed datasets can lead to suboptimal model performance and is not a recommended practice in data preparation.
- D. Correct.
Correct: Data augmentation can help increase the diversity of the image dataset, which often improves the generalization capability of machine learning models.
- E. Incorrect.
Incorrect: Generating synthetic metadata might introduce inaccuracies and is not a standard practice for addressing missing metadata in image datasets.