Research map / Architectures and efficiency / RNNs and normalization
Self-Improving Sparse Retrieval Through Heuristic Representation Refinement and Representation-Focused Learning
Xiaojing Li, Bin Wang, Xiaochun Yang, Meng Luo
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 irrelevant-token removal (ablation) | SIRE (SPLADE++) · MS MARCO dev | MRR@10: 40.7 -> 42.1 | Improved |
| add irrelevant-token removal (ablation) | SIRE (SPLADE++) · BEIR | NDCG@10: 49.0 -> 51.3 | Improved |
| add low-impact token reduction (ablation) | SIRE (SPLADE++) · BEIR | NDCG@10: 50.2 -> 51.3 | Improved |
| add relevant-token augmentation with LLM docs (ablation) | SIRE (SPLADE++) · BEIR | NDCG@10: 49.4 -> 51.3 | Improved |
| add query KL representation loss (ablation) | SIRE (SPLADE++) · MS MARCO dev | MRR@10: 40.2 -> 42.1 | Improved |
| add document contrastive loss (ablation) | SIRE (SPLADE++) · MS MARCO dev | MRR@10: 40.6 -> 42.1 | Improved |
| keep retrieval loss alongside representation losses (ablation) | SIRE (SPLADE++) · MS MARCO dev | MRR@10: 38.8 -> 42.1 | Improved |
| iterate refinement and learning up to 3 rounds (ablation) | SIRE (SPLADE++) · BEIR | NDCG@10: 49.4 -> 51.3 | Improved |
| iterate refinement and learning (ablation) | SIRE (SPLADE++) · MS MARCO dev | MRR@10: 40.9 -> 42.1 | Improved |
| apply heuristic refinement to SPLADE queries at inference | SPLADE · BEIR | NDCG@10: 43.2 -> 44.9 | Improved |
| apply heuristic refinement to SPLADE-v3 queries | SPLADE-v3 · BEIR | NDCG@10: 48.2 -> 50.9 | Improved |
| apply heuristic refinement to SPLADE++ queries | SPLADE++ · MS MARCO dev | MRR@10: 38.0 -> 39.3 | Improved |
Closest papers by meaning
- Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation (ICML 2025 · similarity 0.87)
- Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection (ICLR 2025 · similarity 0.86)
- Retrieval-based Disentangled Representation Learning with Natural Language Supervision (ICLR 2024 · similarity 0.86)
- Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps (NeurIPS 2026 · similarity 0.85)
- Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT (ICML 2024 · similarity 0.85)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (ICLR 2025 · similarity 0.85)
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.
