Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Incorporating Token Importance in Multi-Vector Retrieval
Archish S, Ankit Garg, Kirankumar Shiragur, Neeraj Kayal
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 static IDF query-token weights to Chamfer | ColBERTv2 late interaction · BEIR (13 datasets) | Recall@10: average gain 1.28%, maximum 3.16% | Improved |
| learn token weights with few-shot data instead of fixed Chamfer | ColBERTv2 late interaction · BEIR (13 datasets) | Recall@10: average gain 3.66%, maximum 14.27% | Improved |
| learn few-shot token weights instead of IDF zero-shot weights | Weighted Chamfer · Climate-FEVER (BM25 rerank) | Recall@10: 0.2886 -> 0.3204 | Improved |
| learn few-shot token weights instead of IDF zero-shot weights | Weighted Chamfer · NQ (BM25 rerank) | Recall@10: 0.7628 -> 0.7806 | Improved |
| add IDF token weights to Chamfer | ColBERTv2 · NQ (BM25 rerank) | MRR@10: 0.5296 -> 0.5172 | Hurt |
| add IDF token weights to Chamfer | ColBERTv2 · Webis-Touche2020 (BM25 rerank) | MRR@10: 0.6174 -> 0.5407 | Hurt |
| add token weights to Chamfer | ColBERTv2 · DBPedia-Entity (BM25 rerank) | MRR@10: -0.80% | Hurt |
| add token weights to Chamfer | ColBERTv2 · DBPedia-Entity (BM25 rerank) | nDCG@10: -0.18% | Hurt |
| add token weights to Chamfer | ColBERTv2 · BEIR (end-to-end) | Recall@10 average: +3.04% | Improved |
| add few-shot token weights to Chamfer | ColBERTv2 · SCIDOCS (BM25 rerank) | Recall@10: 0.1575 -> 0.1655 | Improved |
| add few-shot token weights to Chamfer | ColBERTv2 · Quora (BM25 rerank) | Recall@10: 0.9319 -> 0.9384 | No clear effect |
| combine two contrastive losses with different negative sets (convex mix) (ablation) | Few-shot Weighted Chamfer · BEIR | Recall@10 | Improved |
Closest papers by meaning
- Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT (ICML 2024 · similarity 0.86)
- MUVERA: Multi-Vector Retrieval via Fixed Dimensional Encoding (NeurIPS 2024 · similarity 0.86)
- Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval (CVPR 2025 · similarity 0.85)
- Hierarchical Retrieval: The Geometry and a Pretrain-Finetune Recipe (NeurIPS 2025 · similarity 0.85)
- BTR: Binary Token Representations for Efficient Retrieval Augmented Language Models (ICLR 2024 · similarity 0.84)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.84)
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.
