NCA-GENM Question 90
Select 3You are tasked with training a generative AI multimodal model that processes both text and image data. The data comes from multiple sources: text documents stored in different file formats (e.g., PDFs, Word documents) and images with inconsistent resolutions and formats. What steps should you take to preprocess this data effectively for model training?
- A
Convert all text data into a uniform format, such as plain text, and clean it by removing unnecessary formatting or special characters.
- B
Resize images to a consistent resolution and normalize pixel values to a common scale.
- C
Directly input all raw data into the model to ensure it learns from diverse, unprocessed examples.
- D
Use Optical Character Recognition (OCR) to extract text from image-based PDFs or scanned documents.
- E
Convert all data into a proprietary format specific to the model framework, even if it introduces significant preprocessing overhead.
Show answer and explanation
Correct answers: A, B, D
Explanation
Preprocessing data from various sources is essential for training generative AI multimodal models. Text should be converted to a uniform, clean format to ensure consistency, while images should be resized and normalized for compatibility. OCR helps extract text from non-editable document types. These steps ensure the model receives well-structured input, reducing noise and improving training efficiency, whereas skipping preprocessing or introducing unnecessary overhead can degrade performance.
- A. Correct.
Converting text data into a uniform format ensures consistency across input data, which is crucial for the model to learn effectively. Cleaning the data removes noise that could negatively impact model performance.
- B. Correct.
Resizing and normalizing images standardize the input for the model, making it easier to train and improving performance.
- C. Incorrect.
Feeding raw, unprocessed data into the model can introduce noise and inconsistencies, making it harder for the model to learn effectively. Preprocessing is a critical step in preparing data for AI models.
- D. Correct.
OCR is necessary to extract text from non-editable or image-based documents, enabling the integration of this data into the training pipeline.
- E. Incorrect.
While converting data into a proprietary format might sometimes be required, it should not add unnecessary preprocessing overhead unless it's truly beneficial for the specific model framework.