Research map / Large language models / Language, safety and interpretability
UniAPO: Unified Multimodal Automated Prompt Optimization
Qipeng Zhu, Yanzhe Chen, Huasong Zhong, Jie Chen, Yan Li, Zhixin Zhang, Junping Zhang, Zhenheng Yang
Read the paperCheck your idea against 137,217 papers
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 UniAPO instead of ERM* for prompt optimization | GPT-4o predictor · LIAR | F1: 65.2 -> 78.7 | Improved |
| use UniAPO instead of ERM* on WebNLG | GPT-4o predictor · WebNLG | ROUGE-L: 52.1 -> 53.2 | Improved |
| use UniAPO instead of ERM* on ETHOS (GPT-4o) | GPT-4o predictor · ETHOS | F1: 95.6 -> 98.1 | Improved |
| use UniAPO instead of ERM* with QwenVL2.5-72B predictor | QwenVL2.5-72B · LIAR | F1: 67.4 -> 73.1 | Improved |
| use UniAPO instead of ERM* on BBH-navigate with Qwen predictor | QwenVL2.5-72B · BBH-navigate | F1: 93.3 -> 95.8 | Improved |
| add both E-step and M-step to simple-prompt baseline (ablation) | UniAPO · Video KE Beauty | F1: 36.7 -> 75.2 | Improved |
| add either E-step or M-step alone (ablation) | UniAPO · Video KE Beauty | F1: 36.7 -> 66.3 / 67.8 | Improved |
| use UniAPO feedback generator instead of ERM*'s (ablation) | UniAPO with ERM* prompt optimizer · Video KE Beauty | F1: 68.3 -> 73.1 | Improved |
| use UniAPO prompt optimizer instead of ERM*'s (ablation) | ERM* feedback generator · Video KE Beauty | F1: 68.3 -> 70.7 | Improved |
| add long-term feedback memory (ablation) | UniAPO · Video KE Beauty | F1: 70.9 -> 74.7 | Improved |
| add long-term prompt memory (ablation) | UniAPO · Video KE Beauty | F1: 71.3 -> 74.7 | Improved |
| add historical prompts and beam size beyond ERM* (ablation) | UniAPO · Video KE (Figure 5b) | F1 | Improved |
Closest papers by meaning
- Multimodal Prompt Optimization: Why Not Leverage Multiple Modalities for MLLMs (ICLR 2026 · similarity 0.93)
- The Devil is in the Prompts: Retrieval-Augmented Prompt Optimization for Text-to-Video Generation (CVPR 2025 · similarity 0.89)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization (NeurIPS 2024 · similarity 0.88)
- Dynamic Cross-Modal Prompt Generation for Multimodal Continual Instruction Tuning (NeurIPS 2026 · similarity 0.88)
- Unleashing the Potential of Large Language Models as Prompt Optimizers: Analogical Analysis with Gradient-based Model Optimizers (AAAI 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.
