Research map / Large language models / Language, safety and interpretability
Training Diffusion Language Models for Black-Box Optimization
Zipeng Sun, Can Chen, Ye Yuan, Haolun Wu, Jiayao Gu, Christopher Pal, Xue Liu
AI-extracted. The measured effects below were read by AI from the paper's full text (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 diffusion LLM (LLaDA) instead of autoregressive LLaMA backbone (ablation) | DiBO · TF Bind 8 | normalized 100th percentile score: 0.915 -> 0.946 | Improved |
| use diffusion LLM instead of autoregressive backbone (ablation) | DiBO · TF Bind 10 | normalized 100th percentile score: 0.682 -> 0.741 | Improved |
| use diffusion LLM instead of autoregressive backbone (ablation) | DiBO · Ant Morphology | normalized 100th percentile score: 0.930 -> 0.932 | Improved |
| use diffusion LLM instead of autoregressive backbone (ablation) | DiBO · D'Kitty Morphology | normalized 100th percentile score: 0.912 -> 0.912 | No clear effect |
| add delimiter tokens (ablation) | DiBO · Ant Morphology (DA stage) | normalized 100th percentile score: 0.693 -> 0.875 | Improved |
| add delimiter tokens (ablation) | DiBO · TF Bind 8 (RL stage) | normalized 100th percentile score: 0.943 -> 0.946 | Improved |
| add RL stage on top of DA+SFT (ablation) | DiBO · TF Bind 8 | normalized 100th percentile score: 0.939 -> 0.946 | Improved |
| add SFT stage to DA+RL (ablation) | DiBO · Ant Morphology | normalized 100th percentile score: 0.858 -> 0.932 | Improved |
| add domain adaptation stage to SFT+RL (ablation) | DiBO · Ant Morphology | normalized 100th percentile score: 0.913 -> 0.932 | Improved |
| add DA and SFT before RL (ablation) | DiBO · TF Bind 8 | normalized 100th percentile score: 0.437 -> 0.946 | Improved |
| add similarity-conditioned context (ablation) | DiBO · TF Bind 8 | normalized 100th percentile score: 0.730 -> 0.946 | Improved |
| add similarity-conditioned context (ablation) | DiBO · Ant Morphology | normalized 100th percentile score: 0.906 -> 0.932 | Improved |
Closest papers by meaning
- Towards Universal Offline Black-Box Optimization via Learning Language Model Embeddings (ICML 2025 · similarity 0.89)
- From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs (NeurIPS 2026 · similarity 0.88)
- Learning Surrogates for Offline Black-Box Optimization via Gradient Matching (ICML 2024 · similarity 0.87)
- BBOWP-Bench: Evaluating LLMs on Black-Box Optimization Word Problems (NeurIPS 2026 · similarity 0.87)
- Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-based Decoding (NeurIPS 2025 · similarity 0.86)
- Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization (NeurIPS 2024 · similarity 0.86)
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.
