Definition
The computational process of retrieving the most semantically similar data points to a query vector within a high-dimensional embedding space, serving as the core retrieval mechanism for RAG systems. It involves a trade-off between exhaustive search accuracy (latency-heavy) and approximate nearest neighbor (ANN) speed (precision-lossy).
In RAG, this refers to vector retrieval in embedding space, not the K-Nearest Neighbors (KNN) classification algorithm used in traditional supervised learning.
"A laser pointer hitting a specific spot on a massive star map, where the closest surrounding stars represent the most relevant pieces of information."
- Vector Embedding(Prerequisite)
- Cosine Similarity(Component)
- ANN (Approximate Nearest Neighbor)(Optimization Technique)
- HNSW (Hierarchical Navigable Small Worlds)(Algorithm)
Conceptual Overview
The computational process of retrieving the most semantically similar data points to a query vector within a high-dimensional embedding space, serving as the core retrieval mechanism for RAG systems. It involves a trade-off between exhaustive search accuracy (latency-heavy) and approximate nearest neighbor (ANN) speed (precision-lossy).
Disambiguation
In RAG, this refers to vector retrieval in embedding space, not the K-Nearest Neighbors (KNN) classification algorithm used in traditional supervised learning.
Visual Analog
A laser pointer hitting a specific spot on a massive star map, where the closest surrounding stars represent the most relevant pieces of information.