NCA-GENM Question 93
Select 3You are tasked with building a multimodal generative AI model that combines text and image data from various sources. During the data preprocessing stage, you notice that the image dataset contains varying resolutions and the text dataset has missing values in some fields. How should you handle these issues to ensure the data is properly managed and prepared for training?
- A
Resize all images to a consistent resolution suitable for the model's requirements.
- B
Remove all text entries with missing values to ensure data quality.
- C
Impute missing text values using a placeholder or by leveraging contextual information.
- D
Normalize pixel values of images to fall within a standard range like 0 to 1.
- E
Combine text and image data in their raw forms without further preprocessing to preserve their originality.
Show answer and explanation
Correct answers: A, C, D
Explanation
Preprocessing multimodal data requires addressing inconsistencies in resolution, missing values, and scaling. Resizing images ensures compatibility with the model architecture, while imputing missing text values retains data usability. Normalizing pixel values is a standard practice to standardize image data for model training. Removing all missing text entries or using raw data without preprocessing can result in data loss or poor model performance.
- A. Correct.
Correct: Resizing images to a consistent resolution is critical to ensure uniformity and compatibility with the model's input layer.
- B. Incorrect.
Incorrect: Removing all text entries with missing values may lead to information loss, which can negatively impact the model's performance.
- C. Correct.
Correct: Imputing missing text values ensures that the dataset remains usable while maintaining as much information as possible.
- D. Correct.
Correct: Normalizing pixel values is a standard preprocessing step to help the model train effectively and achieve better convergence.
- E. Incorrect.
Incorrect: Combining data in raw form without preprocessing can lead to inconsistencies and poor model performance.