# When Does Dense Retrieval Need Asymmetric Geometry? A Bias-Variance Theory of Shared and Dual Projections

arXiv: https://arxiv.org/abs/2609.32488 · HTML: https://mlsanity.com/paper/2609.32488/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Generation and context integration, one of 15 directions in Retrieval-augmented generation.
- Builds on 21 of its 41 references that are in the map.
- First posted on arXiv 26 Sep 2026.

## Placed in

- Retrieval-augmented generation / [Generation and context integration](https://mlsanity.com/rag/generation-and-context-integration/)

## Builds on (most cited first, 21 of 21)

- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Dense Passage Retrieval for Open-Domain Question Answering](https://mlsanity.com/paper/2004.04906/) (2020)
- [Text Embeddings by Weakly-Supervised Contrastive Pre-training](https://mlsanity.com/paper/2212.03533/) (2022)
- [BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models](https://mlsanity.com/paper/2104.08663/) (2021)
- [Unsupervised Dense Information Retrieval with Contrastive Learning](https://mlsanity.com/paper/2112.09118/) (2021)
- [Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval](https://mlsanity.com/paper/2007.00808/) (2020)
- [Towards General Text Embeddings with Multi-stage Contrastive Learning](https://mlsanity.com/paper/2308.03281/) (2023)
- [C-Pack: Packed Resources For General Chinese Embeddings](https://mlsanity.com/paper/2309.07597/) (2023)
- [Precise Zero-Shot Dense Retrieval without Relevance Labels](https://mlsanity.com/paper/2212.10496/) (2022)
- [RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering](https://mlsanity.com/paper/2010.08191/) (2020)
- [GPL: Generative Pseudo Labeling for Unsupervised Domain Adaptation of Dense Retrieval](https://mlsanity.com/paper/2112.07577/) (2021)
- [Rate-Optimal Perturbation Bounds for Singular Subspaces with Applications to High-Dimensional Statistics](https://mlsanity.com/paper/1605.00353/) (2016)
- [COCO-DR: Combating Distribution Shifts in Zero-Shot Dense Retrieval with Contrastive and Distributionally Robust Learning](https://mlsanity.com/paper/2210.15212/) (2022)
- [Exploring Dual Encoder Architectures for Question Answering](https://mlsanity.com/paper/2204.07120/) (2022)
- [Matryoshka-Adaptor: Unsupervised and Supervised Tuning for Smaller Embedding Dimensions](https://mlsanity.com/paper/2407.20243/) (2024)
- [More Robust Dense Retrieval with Contrastive Dual Learning](https://mlsanity.com/paper/2107.07773/) (2021)
- [Search-Adaptor: Embedding Customization for Information Retrieval](https://mlsanity.com/paper/2310.08750/) (2023)
- [Stochastic Canonical Correlation Analysis](https://mlsanity.com/paper/1702.06533/) (2017)
- [End-to-End Cross-Modality Retrieval with CCA Projections and Pairwise Ranking Loss](https://mlsanity.com/paper/1705.06979/) (2017)
- [Computing the degrees of freedom of rank-regularized estimators and cousins](https://mlsanity.com/paper/1909.10143/) (2019)
- [Align Then Adapt: Label-Efficient Adapter Learning for Asymmetric Dense Retrieval](https://mlsanity.com/paper/2604.03403/) (2026)

## Abstract (arXiv)

Dense retrieval powers retrieval-augmented generation, semantic search, and question answering, yet the theoretical basis for choosing between shared and dual query-document projections remains unclear. We introduce a bias-variance theory for low-rank bilinear scoring. Shared projections induce positive-semidefinite operators, whereas dual projections realize arbitrary low-rank operators. We derive their exact approximation gap and prove a local Gaussian boundary: dual has lower risk exactly when squared directional signal exceeds the estimation cost of its additional degrees of freedom. This boundary motivates the Cross-fitted Asymmetry Risk Selector (CARS), which estimates reproducible directional signal from training pairs; its Gaussian counterpart admits exact selection-power and regret formulas. Guided by the theory, we run retrieval experiments across multiple datasets and embedding models. The mean Dual-minus-Shared NDCG@10 advantage more than doubles as query rotation increases from 0 degrees to 90 degrees. In the rank-sample-size grids, Shared wins 13 of 16 cells at n=32, whereas Dual wins all 32 cells at n=1024 and n=2048. Consistent with this shift, all 168 comparable operator-risk curves move toward Dual as training data grow. Compared to the two fixed-geometry baselines, CARS reduces held-out regret by 49-96% and achieves 90.1% mean geometry-selection accuracy.