NCA-AIIO Question 70
Select 3You are assisting a senior team member in analyzing a dataset for training an AI model. The dataset contains images stored in an unstructured format, and some of the images are corrupted. Your task is to identify and remove these corrupted files as part of the preprocessing step. Which of the following actions should you take under the supervision of the senior team member?
- A
Write a script to scan the dataset and identify corrupted images by attempting to load them.
- B
Remove corrupted files manually by reviewing individual images one by one.
- C
Document the process of identifying corrupted files and share it with the senior team member for review.
- D
Use a third-party library or tool to automate the detection of corrupted images.
- E
Skip the corrupted files and proceed with training the model as they are unlikely to affect performance.
Show answer and explanation
Correct answers: A, C, D
Explanation
Identifying and removing corrupted images is an essential part of data preprocessing in AI workflows. Writing a script or using automated tools ensures efficiency and accuracy, while documenting the process and obtaining supervision from a senior team member ensures adherence to best practices. Manually reviewing images is impractical, and skipping corrupted files can lead to suboptimal model performance.
- A. Correct.
Writing a script to scan the dataset and identify corrupted images is a practical and efficient approach under supervision, as it automates the detection process.
- B. Incorrect.
Manually reviewing each image is highly inefficient and prone to human error, especially for large datasets, making it an impractical choice.
- C. Correct.
Documenting the process and sharing it with a senior team member ensures proper supervision and validation of the steps taken, which aligns with best practices.
- D. Correct.
Using a third-party library or tool to automate the detection is a reliable and time-saving method that can be employed under supervision.
- E. Incorrect.
Skipping corrupted files without addressing them is not a recommended practice, as it could negatively impact the model's performance or introduce bias in the dataset.