Research map / Image generation / Autoregressive and latent generation
BANG: Bridging Autoregressive and Non-autoregressive Generation with Large Scale Pretraining
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 BANG pretraining for semi-NAR generation rather than the strongest semi-NAR baseline | SQuAD 1.1 question generation · SQuAD 1.1 | OVERALL: 14.89 -> 28.90 | Helped |
| use BANG pretraining for NAR generation rather than the strongest NAR baseline | SQuAD 1.1 question generation · SQuAD 1.1 | OVERALL: 14.54 -> 25.27 | Helped |
| use BANG pretraining for semi-NAR generation rather than the strongest semi-NAR baseline | XSum summarization · XSum | OVERALL: 19.95 -> 25.19 | Helped |
| use BANG pretraining for NAR generation rather than the strongest NAR baseline | XSum summarization · XSum | OVERALL: 16.60 -> 22.99 | Helped |
| use BANG pretraining for semi-NAR generation rather than the strongest semi-NAR baseline | PersonaChat dialogue generation · PersonaChat | OVERALL: 20.12 -> 21.66 | Helped |
| use BANG pretraining for NAR generation rather than the strongest NAR baseline | PersonaChat dialogue generation · PersonaChat | OVERALL: 14.15 -> 20.05 | Helped |
| add AR-NAR bridging objectives to NAR-only pretraining (ablation) | BANG NAR finetuning · SQuAD 1.1 | ROUGE-L: 41.15 -> 44.07 | Helped |
| add AR-NAR bridging objectives to NAR-only pretraining (ablation) | BANG NAR finetuning · XSum | ROUGE-1: 30.43 -> 32.50 | Helped |
| use AR (MASS) pretraining instead of NAR-only pretraining (ablation) | NAR finetuning · SQuAD 1.1 | ROUGE-L: 41.15 -> 40.20 | Hurt |
| use AR (MASS) pretraining instead of NAR-only pretraining (ablation) | NAR finetuning · XSum | ROUGE-1: 30.43 -> 26.10 | Hurt |
Closest papers by meaning
- Beyond Autoregression: Fast LLMs via Self-Distillation Through Time (ICLR 2025 · similarity 0.86)
- Mask-Enhanced Autoregressive Prediction: Pay Less Attention to Learn More (ICML 2025 · similarity 0.84)
- MARS: Enabling Autoregressive Models Multi-Token Generation (NeurIPS 2026 · similarity 0.84)
- Reusing Pre-Training Data at Test Time is a Compute Multiplier (ICLR 2026 · similarity 0.84)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.84)
- Silencer: From Discovery to Mitigation of Self-Bias in LLM-as-Benchmark-Generator (NeurIPS 2025 · 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.