Research map / Large language models / Retrieval and knowledge
Autoregressive Search Engines: Generating Substrings as Document Identifiers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 FM-index frequency (LM+FM) scoring over LM-only scoring (ablation) | SEAL · NQ320k | hits@10: 68.7 -> 72.0 | Helped |
| add intersective multi-ngram scoring (ablation) | SEAL · NQ320k | hits@10: 72.0 -> 74.5 | Helped |
| add intersective multi-ngram scoring (ablation) | SEAL · Natural Questions test | accuracy@100: 81.1 -> 86.3 | Helped |
| add FM-index frequency scoring (ablation) | SEAL · Natural Questions test | accuracy@100: 73.1 -> 81.1 | Helped |
| add intersective scoring (ablation) | SEAL · KILT dev passage-level | R-precision: 27.6 -> 54.5 | Helped |
| use FM-index constrained decoding (ablation) | SEAL intersective · Natural Questions test | accuracy@100: 86.2 -> 86.3 | No clear effect |
| increase beam size from 3 to 15 (ablation) | SEAL intersective · Natural Questions test | accuracy@100: 84.9 -> 86.3 | Helped |
| increase beam size from 10 to 15 (ablation) | SEAL intersective · Natural Questions test | accuracy@100: 86.4 -> 86.3 | No clear effect |
| add unsupervised training examples (ablation) | SEAL intersective · Natural Questions test | accuracy@100: 85.4 -> 86.3 | Helped |
| SEAL vs DPR on passage retrieval | SEAL intersective · KILT dev passage-level | R-precision: 42.5 -> 54.5 | Helped |
| SEAL vs DPR top-20 retrieval | SEAL intersective · Natural Questions test | accuracy@20: 80.1 -> 76.2 | Hurt |
| longer ngram length for LM scoring (3 to 5) (ablation) | SEAL (LM) · NQ320k | hits@10: 66.5 -> 68.7 | Helped |
Closest papers by meaning
- Learning to Rank in Generative Retrieval (AAAI 2024 · similarity 0.87)
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.87)
- Beyond Autoregression: Fast LLMs via Self-Distillation Through Time (ICLR 2025 · similarity 0.86)
- Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankers (ICLR 2025 · similarity 0.86)
- Mask-Enhanced Autoregressive Prediction: Pay Less Attention to Learn More (ICML 2025 · similarity 0.85)
- Making Retrieval-Augmented Language Models Robust to Irrelevant Context (ICLR 2024 · 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.