フリー問題

NVIDIA-Certified Associate: Generative AI Multimodal のフリー問題 12 / 20 問目

問題文

A team must sort incoming product photographs into 40 categories that the merchandising team defined last week. No labeled examples exist for any of them, and the list will keep changing. Which approach can start working today, and what does it depend on?

選択肢

  1. Embed each category name as text with an aligned image-and-text model and assign each photograph to its nearest category name; the quality depends on how well the names describe what is visible.
  2. Fine-tune an image classifier on a small number of examples per category collected by hand, and a model that has seen the actual categories will always separate them better.
  3. Match the category names against the product descriptions with keyword search.
  4. Cluster the photographs by their embeddings and assign the 40 names to the resulting groups.

解答・解説を確認するには

正解と解説の確認、回答の記録には無料登録が必要です。登録すると演習モードでフリー問題に回答し、正誤と解説をその場で確認できます。