Research map / Large language models / Instruction and fine-tuning
SMILE: Extended Deep Submodular Function-Based Instruction and In-context Learning Demonstration Selection
Zihan Chen, Chengshuai Shi, Song Wang, Jundong Li, Cong Shen
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 |
|---|---|---|---|
| add query-relevance expert f_SQ (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 58.5 -> 63.0 | Improved |
| add query-instruction scorer s_QI (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 58.0 -> 63.0 | Improved |
| add diversity expert f_SS (ablation) | SMILE (Qwen3-4B-Instruct) · Date | accuracy: 44.8 -> 63.2 | Improved |
| add diversity expert f_SS (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 59.1 -> 63.0 | Improved |
| add instruction-compatibility expert f_SI (ablation) | SMILE (Qwen3-4B-Instruct) · XSum | ROUGE-L F1: 17.6 -> 18.5 | Improved |
| add instruction-compatibility expert f_SI (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 61.8 -> 63.0 | Improved |
| use hard min aggregation instead of mean (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 58.5 -> 63.0 | Improved |
| use hard min at inference instead of softmin (ablation) | SMILE (Qwen3-4B-Instruct) · six-dataset average | average score: 61.0 -> 63.0 | Improved |
| joint instruction and demonstration selection (SMILE) instead of APE+R.S. | prompt optimization (Qwen3-4B-Instruct) · six-dataset average | average score: 59.8 -> 63.0 | Improved |
| joint instruction and demonstration selection (SMILE) instead of APE+R.S. | prompt optimization (Llama3.1-8B-Instruct) · six-dataset average | average score: 57.4 -> 60.7 | Improved |
| transfer surrogate trained on Qwen feedback to Gemini2.5-Flash | SMILE-Q · five-dataset average | average score: 78.8 -> 83.7 | Improved |
| SMILE-L vs strongest baseline APE+R.S. on Gemini2.5-Flash | SMILE-L · five-dataset average | average score: 83.1 -> 83.8 | Improved |
Closest papers by meaning
- Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization (NeurIPS 2024 · similarity 0.89)
- SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models (ICLR 2025 · similarity 0.88)
- Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective (ICML 2025 · similarity 0.87)
- Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models (ICLR 2025 · similarity 0.87)
- Use Your INSTINCT: INSTruction optimization for LLMs usIng Neural bandits Coupled with Transformers (ICML 2024 · similarity 0.87)
- InstructZero: Efficient Instruction Optimization for Black-Box Large Language Models (ICML 2024 · similarity 0.87)
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.
