Approaches to Data Labelling CT-AI Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   An autonomous vehicle company needs to label video frames to train a model that identifies pedestrians and their bounding boxes. The labelling team is experienced with image classification but new to this task. What type of labelling is required, and what is a key challenge they will face?

   • AObject detection; the challenge is drawing accurate bounding boxes around pedestrians in each frame.
   • BImage classification; the challenge is distinguishing between different pedestrian poses.
   • CInstance segmentation; the challenge is separating overlapping pedestrians.
   • DSemantic segmentation; the challenge is labelling every pixel of the pedestrian.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A team is labelling customer support tickets for a model that classifies them into categories (billing, technical, etc.). Two annotators label the same set of 100 tickets and agree on 80 of them. What should the team do to improve labelling consistency?

   • AReview the 20 disagreements, refine the labelling guidelines, and retrain the annotators on the ambiguous cases.
   • BHave a third annotator label only the 20 disagreements and use their labels as the ground truth.
   • CDiscard the 20 disagreements and use only the 80 agreed-upon tickets for training.
   • DUse a majority vote among the two annotators and a third automated system to decide the 20 disagreements.
   Reveal Solution Discussion Select an answer first
4. 3foundation · medium

   Why is data protection important when labelling datasets that contain personal information?

   • ATo comply with regulations like GDPR and prevent unauthorized access to sensitive data
   • BTo make the labelling process faster
   • CTo ensure the labels are accurate
   • DTo reduce the size of the dataset
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A team is labelling a large dataset of satellite images to identify different land cover types (forest, water, urban). They have a limited budget and need to label quickly. They decide to use a model-based automated labelling approach. What is the most important trade-off they should consider?

   • AAutomated labelling is fast and cheap, but may produce errors that require human review to correct.
   • BAutomated labelling eliminates the need for labelling guidelines.
   • CAutomated labelling is more accurate than manual labelling, so no human review is needed.
   • DAutomated labelling is only suitable for text data, not images.
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A team is labelling a large dataset of text documents for a legal document classification model. They need to ensure high accuracy and consistency. They are evaluating different labelling platforms. What is the most important feature to look for in a platform for this task?

   • AIntegration with a crowdsourcing platform to access a large workforce.
   • BSupport for custom labelling schemas and the ability to define complex rules for validation.
   • CSupport for image annotation with bounding boxes and polygons.
   • DPre-built models for automatic labelling that eliminate the need for manual labelling.
   Reveal Solution Discussion Select an answer first