Identify Features of Generative AI Solutions AI-900 Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A music streaming service wants to create personalized playlists that include new, AI-generated songs in the style of each user's favorite artists. Which approach should they use?

   • AUse a classification model to tag songs by genre.
   • BTrain a generative model on the audio features of existing songs to compose new tracks.
   • CUse a recommendation algorithm to reorder existing songs in the catalog.
   • DUse a clustering model to group similar artists.
   Reveal Solution Discussion Select an answer first
2. 32foundation · easy

   How can generative AI be used in simulation for autonomous vehicle training?

   • ABy analyzing sensor data to detect lane markings
   • BBy predicting traffic congestion in real time
   • CBy optimizing the vehicle's engine performance
   • DBy generating realistic driving scenarios that are rare in real-world data
   Reveal Solution Discussion Select an answer first
4. 33foundation · easy

   Which scenario is a common use case for generative AI?

   • AClustering customers into segments based on purchasing behavior
   • BGenerating synthetic customer reviews to train a sentiment analysis model
   • CPredicting the likelihood of a patient developing a disease
   • DDetecting fraudulent transactions in a financial dataset
   Reveal Solution Discussion Select an answer first
5. 34expert · hard

   A research team is building a generative AI model to create synthetic medical images for training a diagnostic model. They have a small dataset of real images that is skewed toward one demographic group. They need to improve the diagnostic model's accuracy for underrepresented groups. Which approach best balances the need for realistic data and fairness?

   • ATrain the diagnostic model only on the original dataset to avoid introducing synthetic bias.
   • BTrain the generative model on the existing data, then generate additional images for the underrepresented groups and validate them with domain experts.
   • CUse a pre-trained generative model without fine-tuning to generate images for all groups.
   • DGenerate a large number of synthetic images for all groups without validation to quickly expand the dataset.
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A robotics company is developing a system that generates realistic synthetic sensor data (e.g., lidar and camera) to train their robots for warehouse navigation. They need high-fidelity data that closely matches real-world sensor noise. They are considering using a GAN or a VAE. What is the key trade-off they should consider?

   • AGANs are easier to train and guarantee diverse outputs.
   • BGANs typically produce sharper, more realistic outputs but are harder to train and may suffer from mode collapse.
   • CVAEs typically produce sharper outputs but are harder to train than GANs.
   • DVAEs are always the best choice for high-fidelity sensor data.
   Reveal Solution Discussion Select an answer first