1Z0-184-25 Question 74
Single answerYour data science team has generated vector embeddings for a large set of product images. They plan to store these embeddings in Oracle Database 23c on Oracle Cloud Infrastructure and need to perform an exact similarity search (exact nearest neighbor) to retrieve the most similar images for each query. Which approach should you recommend to ensure accurate and efficient results?
- A
Store the embeddings in a vector column and create an Exact Vector Index for similarity queries.
- B
Use approximate vector indexes to speed up similarity search, then filter results manually.
- C
Store embeddings as unindexed BLOBs and use a custom PL/SQL function to compare vectors.
- D
Rely on a traditional full-text index to perform exact similarity matching on the image embeddings.
Show answer and explanation
Correct answer: A
Explanation
To perform an exact similarity search in Oracle Database 23c, you can leverage the built-in vector data type and create an Exact Vector Index. This specialized index supports distance-based queries on vector columns without the need for custom code or external filters. Refer to the official Oracle Database 23c documentation on vector search for details on index creation and best practices.
- A. Correct.
Correct. Oracle Database 23c supports a vector data type and indexing strategies specifically designed for similarity search. By creating an Exact Vector Index on the vector column, you can achieve precise comparisons without needing a post-processing filter pass. This is the recommended approach for workloads that require full precision in similarity matches.
- B. Incorrect.
Incorrect. Approximate indexes (e.g., approximate nearest neighbor) can improve performance but do not provide an exact result set without additional filtering or scoring. While they can be useful for large datasets with real-time requirements, this approach does not guarantee fully accurate matches unless combined with a secondary pass.
- C. Incorrect.
Incorrect. Storing vectors as unindexed BLOBs would require a full table scan or custom function for every query, leading to poor performance and a more complex implementation. You lose the benefits of specialized vector indexing that is built into Oracle Database 23c.
- D. Incorrect.
Incorrect. A full-text index is designed for textual document searching rather than vector-based embeddings. It does not natively support distance metrics or dot-product comparisons required for similarity search on numerical vectors. Therefore, it cannot provide an accurate or efficient exact similarity match on image embeddings.