1Z0-184-25 Question 38
Select 2You are designing an image recognition system on Oracle Cloud Infrastructure, storing feature embeddings in an Autonomous Database table using the VECTOR(256) type. You need to efficiently perform similarity searches (e.g., find the closest matching embeddings) in real time. Which TWO actions should you take to ensure optimal performance and accuracy when creating and using a vector index?
- A
Store embeddings in a VECTOR column with the same dimension as your model output and create a specialized vector index for approximate nearest neighbor searches.
- B
Partition the VECTOR column by each of the 256 dimensions to reduce overhead and rely on B-Tree indexing for similarity queries.
- C
Use a standard B-Tree index on the VECTOR column to leverage existing index optimizations for nearest neighbor queries.
- D
Normalize or scale the embedding values according to your ML model� scale before creating the vector index, ensuring consistent distance metrics.
Show answer and explanation
Correct answers: A, D
Explanation
When dealing with high-dimensional embeddings in Oracle Autonomous Database, a specialized vector index significantly improves similarity search performance. Ensuring that the dimension of the VECTOR column matches your model output and normalizing embeddings before indexing leads to more accurate and efficient queries. For more details, refer to the Oracle Database 23c documentation on Vector Search Indexes and Oracle Cloud Infrastructure best practices for machine learning data storage.
- A. Correct.
Correct. Storing data in an appropriately sized VECTOR column and creating a vector index (often an approximate nearest neighbor index) is the recommended approach for efficient high-dimensional similarity searches. Oracle� specialized vector index is optimized for such queries.
- B. Incorrect.
Incorrect. Partitioning by each dimension is not a standard best practice for vector data and would complicate data management without improving similarity search performance. Specialized vector indexing is more efficient than manual partitioning by dimension.
- C. Incorrect.
Incorrect. B-Tree indexes are optimized for exact match or range queries on scalar data, not for similarity calculations on high-dimensional embeddings. They are not suitable to achieve the performance requirements of vector-based searches.
- D. Correct.
Correct. Normalizing or scaling embeddings ensures that distance metrics (e.g., cosine similarity, Euclidean distance) operate consistently across data points, improving accuracy and performance when using the vector index.