1Z0-184-25 Question 46
Select 2Your team is developing an image similarity application using embeddings stored in an Oracle Autonomous Database table. You plan to create vector indexes on the embeddings column to speed up AI vector search across millions of images. Which two steps or best practices should you implement to optimize vector search performance in this scenario?
- A
Define the vector column using the exact dimension of your embeddings (e.g., VECTOR(512) if you have 512-dimensional embeddings).
- B
Use the approximate nearest neighbor indexing feature optimized for high-dimensional data to significantly reduce query latency.
- C
Rely on a conventional B-Tree index for vector columns, as it offers the quickest approach for high-dimensional data searches.
- D
Disable table partitioning because it inherently conflicts with vector indexing and impairs query performance.
Show answer and explanation
Correct answers: A, B
Explanation
When creating vector indexes in Oracle Autonomous Database (or Oracle Database 23c and later versions), you must match the column� vector dimension to your embedding size and use an appropriate indexing method�often approximate nearest neighbor�for optimal performance. Conventional indexes like B-Tree are not suited for high-dimensional vector data. Partitioning strategies can still be leveraged with vector indexing. For more details, refer to Oracle� documentation on vector data types and index creation for AI workloads.
- A. Correct.
Option 1 is correct. It is crucial that the vector column� defined dimension matches the size of your embeddings. For instance, if your embeddings are 512-dimensional, you must configure the column as VECTOR(512) to ensure the index can properly handle all components.
- B. Correct.
Option 2 is correct. For large-scale AI workloads, approximate nearest neighbor indexing is specifically designed to handle high-dimensional embeddings efficiently and reduce search times compared to a brute-force or purely exact approach.
- C. Incorrect.
Option 3 is incorrect. B-Tree indexes are typically used for scalar data and are not well-suited for high-dimensional vector searches; they lead to poor performance as dimensionality grows. Specialized vector indexes are much more efficient.
- D. Incorrect.
Option 4 is incorrect. There is no general requirement to disable table partitioning for vector indexes. Partitioning can still be beneficial in managing data distribution and query performance, especially if carefully planned alongside vector indexes.