NCA-GENM Question 54
Select 3You are tasked with preparing a dataset for a multimodal generative AI model that combines text and image inputs. The dataset contains missing values in the text descriptions, mislabeled image categories, and inconsistent formatting for numerical metadata. What steps should you prioritize to ensure the data is ready for training?
- A
Identify and resolve missing values in the text descriptions using appropriate imputation techniques.
- B
Standardize the numerical metadata to ensure consistent formatting and scaling.
- C
Ignore mislabeled image categories, as the model can learn to correct for this during training.
- D
Perform exploratory data analysis (EDA) to detect patterns, anomalies, and outliers.
- E
Augment the dataset by generating synthetic data for text and images without addressing the identified issues.
Show answer and explanation
Correct answers: A, B, D
Explanation
To prepare a dataset for a multimodal generative AI model, it is critical to address missing data, standardize numerical metadata, and perform exploratory data analysis to identify and resolve potential issues. Ignoring mislabeled categories or augmenting the data without first ensuring its integrity will likely harm the model's performance.
- A. Correct.
Correct. Resolving missing values is critical for ensuring that the model has sufficient and consistent input data to learn effectively.
- B. Correct.
Correct. Standardizing numerical metadata is essential to avoid introducing biases in the training process and to ensure uniform interpretation by the model.
- C. Incorrect.
Incorrect. Ignoring mislabeled image categories can lead to noisy labels, which will negatively impact the model's ability to learn accurate patterns.
- D. Correct.
Correct. Performing exploratory data analysis is a foundational step to understand the dataset, identify issues, and guide cleaning and transformation efforts.
- E. Incorrect.
Incorrect. Augmenting the dataset without addressing underlying data quality issues will propagate errors and reduce the overall effectiveness of the model.