Definition
The strategic combination of multiple retrieval models or generation outputs to produce a result that is more accurate and robust than any individual model could achieve alone. In RAG pipelines, this typically involves merging results from diverse sources like dense vector search and sparse keyword search using fusion algorithms.
Distinguish from 'Multi-Agent Systems'—Ensemble Methods focus on merging outputs from parallel processes, whereas Multi-Agent systems focus on sequential task delegation.
"A panel of specialists (e.g., a historian and a data scientist) voting on the most relevant document to ensure no single perspective is missed."
- Reciprocal Rank Fusion (RRF)(Component)
- Hybrid Search(Prerequisite)
- Reranking(Component)
- Mixture of Experts (MoE)(Architectural Peer)
Conceptual Overview
The strategic combination of multiple retrieval models or generation outputs to produce a result that is more accurate and robust than any individual model could achieve alone. In RAG pipelines, this typically involves merging results from diverse sources like dense vector search and sparse keyword search using fusion algorithms.
Disambiguation
Distinguish from 'Multi-Agent Systems'—Ensemble Methods focus on merging outputs from parallel processes, whereas Multi-Agent systems focus on sequential task delegation.
Visual Analog
A panel of specialists (e.g., a historian and a data scientist) voting on the most relevant document to ensure no single perspective is missed.