Research map / Architectures and efficiency / RNNs and normalization
No More K-means: Single-Stage Sparse Coding for Efficient Multi-Vector Retrieval
Lixuan Guo, Yifei Wang, Tiansheng Wen, Aosong Feng, Stefanie Jegelka, Chenyu You
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add SSR++ coarse-to-fine pruning (ablation) | SSR · MS MARCO passage subset | retrieval latency (ms): 38.6 -> 17.5 | Improved |
| add SSR++ coarse-to-fine pruning (ablation) | SSR · MS MARCO passage subset | candidates hit: 54,278 -> 3,196 | Improved |
| add SSR++ coarse-to-fine pruning (ablation) | SSR · MS MARCO passage subset | nDCG@10: 45.3 -> 45.2 | No clear effect |
| use SSR sparse late interaction instead of Splade-v3 | SSR-CLS · BEIR 13 datasets | nDCG@10: 51.2 -> 53.4 | Improved |
| use SSR-tok instead of ColBERTv2 for retrieval | SSR-tok · MS MARCO passage | retrieval latency (ms): 56.9 -> 17.5 | Improved |
| apply SSR to frozen Llama-embed-nemotron-8b | SSR-CLS · BEIR 13 datasets (avg) | nDCG@10: 65.8 -> 67.1 | Improved |
| use SSR instead of a linear projector on frozen backbone (ablation) | Llama-8B · BEIR 13 datasets (avg) | nDCG@10: 66.1 -> 67.1 | Improved |
| add a linear projector to frozen e5-mistral-7b (ablation) | e5-mistral-7b · BEIR 13 datasets (avg) | nDCG@10: 58.8 -> 60.2 | Improved |
| use SSR instead of ColBERTv2 on LoTTE search queries | SSR · LoTTE Search (avg) | Success@5: 71.3 -> 73.8 | Improved |
| use SSR instead of ColBERTv2 on LoTTE forum queries | SSR · LoTTE Forum (avg) | Success@5: 65.1 -> 67.3 | Improved |
| increase hidden dimension h from 2^12 to 2^16 (ablation) | SSR · BEIR | nDCG@10 | Mixed |
| reduce sparsity K below 16 (ablation) | SSR · BEIR | nDCG@10 | Hurt |
Closest papers by meaning
- MUVERA: Multi-Vector Retrieval via Fixed Dimensional Encoding (NeurIPS 2024 · similarity 0.88)
- Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation (ICML 2025 · similarity 0.87)
- Adaptive Retrieval and Scalable Indexing for k-NN Search with Cross-Encoders (ICLR 2024 · similarity 0.87)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.86)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.86)
- Your Embedding Model Is SMARTer Than You Think (NeurIPS 2026 · similarity 0.86)
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.
