AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

HypRAG: Hyperbolic Dense Retrieval for Retrieval Augmented Generation

Hiren Madhu, Ngoc Bui, Ali Maatouk, Leandros Tassiulas, Smita Krishnaswamy, Menglin Yang, Sukanta Ganguly, Kiran Srinivasan, Rex Ying

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Language model pretraining and induction · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add OEM pooling instead of CLS pooling (ablation)HyTE-FH · MTEBmean over tasks: 49.33 -> 56.41Improved
add OEM pooling instead of Einstein midpoint (ablation)HyTE-FH · MTEBmean over tasks: 50.33 -> 56.41Improved
add OEM pooling instead of naive mean pooling (ablation)HyTE-FH · MTEBmean over tasks: 47.07 -> 56.41Improved
use fully hyperbolic model instead of Euclidean equivalentEucBERT · MTEBmean over tasks: 54.11 -> 56.41Improved
use fully hyperbolic model instead of Euclidean equivalentEucBERT · RAGBench averagefaithfulness: 0.596 -> 0.732Improved
use hybrid hyperbolic model initialized from EucBERTEucBERT · RAGBench averagefaithfulness: 0.596 -> 0.706Improved
use hyperbolic HyTE-FH instead of EucBERT with HNSWEucBERT · MS MARCO ANNRecall@50: 0.795 -> 0.859Improved
use hyperbolic HyTE-FH instead of EucBERT in ANN searchEucBERT · MS MARCO ANNlatency: 0.29 -> 0.09 msImproved
use hyperbolic HyTE-FH instead of EucBERTEucBERT · MeSH hierarchy-sensitive retrievalRecall@1: 0.810 -> 0.963Improved
use hyperbolic HyTE-FH instead of EucBERTEucBERT · MeSH hierarchy-sensitive retrievalSpecHR: 0.883 -> 0.970Improved
use HyTE-H(Qwen) instead of Qwen-EmbeddingQwen-Embedding · RAGBench averagefaithfulness: 0.784 -> 0.803Improved
use hyperbolic fine-tuning on ModernBERT instead of Euclidean embedders on Emanual/ExpertQAHyTE-H (bert) · RAGBench per-datasetfaithfulnessMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy