1Z0-184-25 Question 44
Single answerYou are building an AI-driven recommendation engine on Oracle Cloud Infrastructure using an Autonomous Database that supports vector queries. Your dataset contains thousands of item embeddings, and you need low-latency searches for the top similar items. Which approach should you use to optimize performance for vector queries in your Autonomous Database environment?
- A
Use a standard B-Tree index on the vector column for nearest neighbor searches
- B
Build a specialized vector index using Oracle� approximate nearest neighbor algorithm on the vector column
- C
Rely on full-text search indices by storing embeddings as text in a JSON column
- D
Partition the vector column by each dimension using range partitions for faster lookups
Show answer and explanation
Correct answer: B
Explanation
When dealing with AI embeddings in Oracle Autonomous Database, using a specialized vector index is essential for performing efficient nearest neighbor searches. Oracle� approximate nearest neighbor algorithms (such as HNSW) are optimized for high-dimensional data and can significantly reduce search latency. For more information, consult the Oracle documentation on vector similarity indexing in Autonomous Database, which outlines best practices for creating and maintaining these specialized indexes in real-world AI/ML workloads.
- A. Incorrect.
Option 1: Incorrect. A B-Tree index is not suitable for high-dimensional vector data and nearest neighbor searches. While B-Tree works well for exact matches or range queries on scalar data, it does not efficiently handle similarity queries on embeddings.
- B. Correct.
Option 2: Correct. A specialized vector index with approximate nearest neighbor capabilities (e.g., HNSW) is designed for high-dimensional vector similarity searches. This index significantly speeds up queries for finding the top similar items in an embedding-based application.
- C. Incorrect.
Option 3: Incorrect. Simply storing embeddings as text in a JSON column and relying on full-text search indices does not leverage approximate nearest neighbor search. It might allow keyword-based lookups, but it does not optimize similarity queries on numeric vector data.
- D. Incorrect.
Option 4: Incorrect. Partitioning by each dimension would be inefficient and overly complex for vector data. Nearest neighbor searches involve comparing embeddings across all dimensions simultaneously, so separate dimension-based partitions would not improve similarity search performance.