A vector index with approximate similarity search is designed specifically for high-performance top-K retrieval over large vector collections. Exact vector search calculates distances against all candidate vectors that satisfy the query predicates, which can become computationally expensive at scale. Approximate nearest-neighbor search uses vector indexing structures to reduce the number of candidate vectors evaluated, significantly improving search latency while accepting a controlled trade-off between performance and recall or accuracy. Oracle AI Database supports vector indexes with organizations such as INMEMORY NEIGHBOR GRAPH and NEIGHBOR PARTITIONS and allows administrators to configure target accuracy.

This requirement explicitly states that approximate top-K results are acceptable, making an approximate vector index the intended architecture. A conventional B-tree index is appropriate for scalar equality, ordering, or range-access patterns, not high-dimensional semantic similarity. JSON Duality Views provide document-relational mapping rather than nearest-neighbor acceleration. Property graph views model entities and relationships and likewise do not serve as vector similarity indexes. The uploaded assessment identifies “a vector index with approximate search” as the correct option.

Study Guide reference: Working with AI and Vector Foundations — vector indexes, approximate nearest-neighbor search, top-K retrieval, and target accuracy.

===============