Definition
A distance metric that measures the number of positions at which two binary vectors differ, specifically used in RAG pipelines to enable ultra-fast similarity searches via Binary Quantization (BQ). While it significantly reduces memory usage and latency compared to Cosine Similarity, it involves a trade-off in retrieval precision due to the loss of floating-point granularity.
Measures bitwise differences in quantized vectors, not character substitutions in strings.
"Comparing two identical rows of light switches and counting how many switches are flipped in opposite directions."
- Binary Quantization(Component)
- Vector Embedding(Prerequisite)
- Cosine Similarity(Alternative Metric)
- Approximate Nearest Neighbor (ANN)(Context)
Conceptual Overview
A distance metric that measures the number of positions at which two binary vectors differ, specifically used in RAG pipelines to enable ultra-fast similarity searches via Binary Quantization (BQ). While it significantly reduces memory usage and latency compared to Cosine Similarity, it involves a trade-off in retrieval precision due to the loss of floating-point granularity.
Disambiguation
Measures bitwise differences in quantized vectors, not character substitutions in strings.
Visual Analog
Comparing two identical rows of light switches and counting how many switches are flipped in opposite directions.