MiniCoil - A Hybrid Sparse Retrieval Model for Scalable and Context-Aware Semantic Search
OpenSource Connections via YouTube
UC San Diego Product Management Certificate — AI-Powered PM Training
Live Online Classes in Design, Coding & AI — Small Classes, Free Retakes
Overview
AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off Coursera Plus Annual.
Unlock All Certificates
Explore MiniCoil, an innovative hybrid sparse retrieval model that combines the interpretability of sparse retrieval with the contextual depth of dense embeddings for scalable semantic search. Learn how this approach achieves high performance with minimal computational overhead by generating compact, sparse representations through transformer-based embeddings and trained, meaning-preserving layers that enable significant dimensionality reduction while retaining semantic information. Discover the hybrid design's dynamic vocabulary expansion capabilities and seamless BM25 fallback for out-of-vocabulary terms, ensuring robust retrieval across diverse datasets. Understand the domain-agnostic architecture that serves as a versatile, general-purpose retrieval solution while supporting fine-tuning for specialized applications in legal and medical search domains. Gain insights into the core architecture, training methodologies, and practical applications for powering search engines, enterprise knowledge systems, and conversational AI, with actionable strategies for developing efficient, context-aware retrieval systems that balance speed, accuracy, and interpretability.
Syllabus
Haystack US 2025 - Thierry Damiba: MiniCoil: A Hybrid Sparse Retrieval Model
Taught by
OpenSource Connections