Research map / Image generation / Text-to-image diffusion
Unsupervised Text Generation by Learning from Search
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 first-stage CE learning from SA results (ablation) | TGLS on Quora · Quora | iBLEU: 14.52 -> 14.97 | Helped |
| Add first-stage CE learning from SA results (ablation) | TGLS on Quora · Quora | Inference time (sec/sample): 5.46 -> 0.06 | Helped |
| Initialize second-stage SA with the GPT2 prediction (ablation) | SA+CE on Quora · Quora | iBLEU: 14.97 -> 15.41 | Helped |
| Use second-stage CE rather than second-stage MM (ablation) | SA+CE+SA on Quora · Quora | iBLEU: 15.70 -> 17.48 | Hurt |
| Use second-stage MM rather than second-stage CE (ablation) | SA+CE+SA on Quora · Quora | BLEU: 21.70 -> 25.00 | Helped |
| Train with MM without first-stage CE warm-starting (ablation) | TGLS on Quora · Quora | generation quality | Hurt |
| Replace SA-only generation with the full search-and-learning framework (ablation) | Paraphrase generation · Quora | iBLEU: 14.52 -> 17.48 | Helped |
| Use TGLS instead of UPSA | Unsupervised paraphrase generation · Quora | iBLEU: 12.02 -> 17.48 | Helped |
| Use TGLS instead of DualRL | Unsupervised formality transfer · GYAFC Family & Relationships | H-mean: 56.15 -> 66.88 | Helped |
| Use unsupervised TGLS instead of supervised LSTM-attn | Formality transfer · GYAFC Family & Relationships | H-mean: 77.34 -> 66.88 | Hurt |
Closest papers by meaning
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.87)
- RetGen: A Joint Framework for Retrieval and Grounded Text Generation Modeling (AAAI 2022 · similarity 0.87)
- UniGen: A Unified Generative Framework for Retrieval and Question Answering with Large Language Models (AAAI 2024 · similarity 0.86)
- SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generation (ICLR 2025 · similarity 0.85)
- The Hyperfitting Phenomenon: Sharpening and Stabilizing LLMs for Open-Ended Text Generation (ICLR 2025 · similarity 0.85)
- Closing the Curious Case of Neural Text Degeneration (ICLR 2024 · 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.