Research map / Large language models / Retrieval and knowledge
Lift Yourself Up: Retrieval-augmented Text Generation with Self-Memory
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 |
|---|---|---|---|
| use self-memory instead of retrieved memory (ablation) | Transformer_uni · JRC-Acquis En->De test | BLEU: 58.58 -> 60.11 | Helped |
| use self-memory instead of retrieved memory (ablation) | Transformer_uni · JRC-Acquis En->Es test | BLEU: 64.12 -> 65.94 | Helped |
| use self-memory instead of retrieved memory (ablation) | XGLM-1.7B in-context learning · En->De | BLEU: 37.87 -> 40.94 | Helped |
| use self-memory instead of kNN demonstrations (ablation) | XGLM-7.5B in-context learning · De->En | BLEU: 55.65 -> 55.12 | No clear effect |
| use kNN demonstrations instead of random demonstrations (ablation) | XGLM-1.7B in-context learning · En->De | BLEU: 11.51 -> 37.87 | Helped |
| use beam output directly as memory (no selector) (ablation) | Transformer_joint retrieval-augmented · JRC-Acquis En->De | BLEU: 58.58 -> 58.43 | No clear effect |
| use self-memory instead of retrieved memory (ablation) | BART_joint · DailyDialog | B-1: 36.72 -> 39.80 | Helped |
| use self-memory instead of retrieved memory (ablation) | BART_joint · DailyDialog | D-1: 6.13 -> 5.84 | Hurt |
| select memory by Distinct score instead of BLEU (ablation) | BART_joint with self-memory · DailyDialog | D-1: 5.84 -> 9.12 | Helped |
| select memory by Distinct score instead of BLEU (ablation) | BART_joint with self-memory · DailyDialog | B-1: 39.80 -> 36.92 | Hurt |
| add self-memory to BART summarizer (ablation) | BART_base · BigPatent | ROUGE-1: +18.5 R1 | Helped |
| add self-memory to BRIO (ablation) | BRIO · XSum | ROUGE-1: +1.2 R1 | Helped |
Closest papers by meaning
- Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR 2024 · similarity 0.90)
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.90)
- Self-Retrieval: End-to-End Information Retrieval with One Large Language Model (NeurIPS 2024 · similarity 0.89)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.88)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.88)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.88)
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.