Research map / Image generation / Autoregressive and latent generation
DiffuSeq: Sequence to Sequence Text Generation with Diffusion Models
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 DiffuSeq instead of Transformer-base | Seq2Seq generation · Open Domain Dialogue | BLEU: 0.0189 -> 0.0139 | Hurt |
| Use DiffuSeq instead of Transformer-base | Seq2Seq generation · Question Generation | BLEU: 0.1663 -> 0.1731 | Helped |
| Use DiffuSeq instead of GPVAE-T5 | Seq2Seq generation · Text Simplification | BLEU: 0.3392 -> 0.3622 | Helped |
| Use DiffuSeq instead of Transformer-base | Seq2Seq generation · Paraphrase | BLEU: 0.2722 -> 0.2413 | Hurt |
| Use DiffuSeq instead of Transformer-base | Seq2Seq generation · Open Domain Dialogue | self-BLEU: 0.3698 -> 0.0144 | Helped |
| Use DiffuSeq instead of GPT2-base FT | Seq2Seq generation · Question Generation | self-BLEU: 0.1403 -> 0.2789 | Hurt |
| Use DiffuSeq instead of GPT2-large FT | Seq2Seq generation · Text Simplification | self-BLEU: 0.6042 -> 0.4642 | Helped |
| Use DiffuSeq instead of Transformer-base | Seq2Seq generation · Paraphrase | self-BLEU: 0.4483 -> 0.2732 | Helped |
| Remove MBR reranking (ablation) | DiffuSeq · Question Generation | BLEU: 0.1567 -> 0.1731 | Hurt |
| Replace jointly trained source embeddings with fixed pretrained BERT-tiny representations (ablation) | DiffuSeq · Question Generation | BLEU: 0.1567 -> 0.0110 | Hurt |
| Replace jointly trained source embeddings with fixed pretrained BERT-tiny representations (ablation) | DiffuSeq · Question Generation | self-BLEU: 0.2789 -> 0.0174 | Helped |
Closest papers by meaning
- Meta-Diffu$B$: A Contextualized Sequence-to-Sequence Text Diffusion Model with Meta-Exploration (NeurIPS 2024 · similarity 0.91)
- Scaling Diffusion Language Models via Adaptation from Autoregressive Models (ICLR 2025 · similarity 0.91)
- Energy-Based Diffusion Language Models for Text Generation (ICLR 2025 · similarity 0.91)
- Theoretical Benefit and Limitation of Diffusion Language Model (NeurIPS 2025 · similarity 0.90)
- Breaking AR’s Sampling Bottleneck: Provable Acceleration via Diffusion Language Models (NeurIPS 2025 · similarity 0.89)
- How to Train Your Latent Diffusion Language Model Jointly With the Latent Space (NeurIPS 2026 · 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.