1Z0-184-25 Question 79
Single answerYou have deployed an image recognition application on Oracle Cloud Infrastructure. Your product catalog images are stored as vector embeddings in an Autonomous Database table using a specialized vector data type. You need to find the most visually similar products to a new image's generated embedding, and your team wants the query results to return quickly even as the product catalog grows. Which approach should you use to achieve fast approximate similarity search using Oracle� vector index capabilities?
- A
Create an approximate L2 vector index on the embedding column and use KNN search with the L2 distance measure to retrieve similar vectors.
- B
Store all embeddings in an Object Storage bucket for direct vector-based lookup using the approximate similarity search functionality.
- C
Convert all embeddings to JSON documents and query them using JSON search operators for approximate similarity.
- D
Rely on a standard B-tree index on the raw numerical values of the embedding column to achieve the fastest approximate similarity search.
Show answer and explanation
Correct answer: A
Explanation
To efficiently perform approximate similarity search in Oracle Autonomous Database, you can leverage the vector indexing feature introduced for high-dimensional data. By creating an approximate vector index on the embedding column and specifying the desired distance measure (e.g., L2 or cosine), you can quickly find the k-nearest neighbors. Refer to Oracle� documentation on vector search in the Autonomous Database to learn how to create the index and run performant KNN queries for large-scale image or text embeddings.
- A. Correct.
Correct. Oracle's vector index functionality supports approximate similarity search by creating an index on the embedding column. Using the L2 distance measure (or other supported metrics) with a KNN-style query will provide fast and scalable approximate similarity results for large datasets.
- B. Incorrect.
Incorrect. Storing vectors in Object Storage alone does not enable direct approximate similarity search. While you can store data offline or for backup, the actual vector indexing and querying must be performed within the database environment that supports vector indexes.
- C. Incorrect.
Incorrect. While JSON can store unstructured data, JSON-based queries are not designed for efficient approximate similarity search on vector embeddings. Oracle� specialized vector indexing provides the required performance and functionality, which goes beyond simple JSON search operators.
- D. Incorrect.
Incorrect. A standard B-tree index is not suited for high-dimensional approximate similarity queries. B-tree indexes are designed for exact matches or range queries on scalar values, not for vector-based KNN similarity queries.