Research map / Large language models / Language, safety and interpretability
Learning More from Less: Unlocking Internal Representations for Benchmark Compression
Yueqi Zhang, Jin Hu, Shaoxiong Feng, Peiwen Yuan, Xinglin Wang, Yiwei Li, Jiayi Shi, Chuyi Tan, Ji Zhang, Boyuan Pan, Yao Hu, Kan Li
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 |
|---|---|---|---|
| cluster on aligned consensus embeddings instead of binary correctness vectors (ablation) | RepCore · five benchmarks, K=30 | Spearman rho: 0.850 -> 0.868 | Improved |
| cluster on aligned consensus embeddings instead of binary correctness vectors (ablation) | RepCore · five benchmarks, K=30 | MAE: 0.055 -> 0.049 | Improved |
| use RepCore instead of AnchorPoints with only five source models | RepCore · five benchmarks, K=30, |S|=5 | Spearman rho: 0.640 -> 0.860 | Improved |
| add PC2 and PC3 to PC1 for extrapolation (ablation) | RepCore · BBH | Spearman rho: K=30: 0.857 -> 0.871 (PC1 vs PC1,2) | Improved |
| use full aligned embedding instead of PC1,2 subspace (ablation) | RepCore · BBH | Spearman rho: K=10: 0.689 -> 0.718; K=50: 0.917 -> 0.913 | Mixed |
| use diverse source pool instead of a single model family (Llama / InternVL) (ablation) | RepCore · five benchmarks, K=30 | Spearman rho: 0.686 -> 0.868 | Improved |
| use diverse source pool instead of strong-only models (ablation) | RepCore · five benchmarks, K=30 | MAE: 0.092 -> 0.049 | Improved |
| increase source pool from 5 to 20 models (ablation) | RepCore · five benchmarks, K=30 | Spearman rho: 0.860 -> 0.864 | No clear effect |
| increase coreset size to K=200 | RepCore vs GP-IRT · five benchmarks | Spearman rho: 0.975 vs 0.973 | No clear effect |
| mask 50% of source hidden states at random (ablation) | RepCore · five benchmarks | Spearman rho: 0.842 -> 0.833 | Hurt |
| mask 50% of most distinct sources' hidden states (ablation) | RepCore · five benchmarks | Spearman rho: 0.842 -> 0.805 | Hurt |
| use RepCore on chronological split (old sources, new targets) | RepCore · five benchmarks, K=30 | Spearman rho: 0.845 -> 0.853 (vs Random) | Improved |
Closest papers by meaning
- When Less is More: The LLM Scaling Paradox in Context Compression (NeurIPS 2026 · similarity 0.90)
- Language Modeling Is Compression (ICLR 2024 · similarity 0.89)
- Compressing LLMs: The Truth is Rarely Pure and Never Simple (ICLR 2024 · similarity 0.89)
- Accuracy is Not All You Need (NeurIPS 2024 · similarity 0.89)
- Can Compressed LLMs Truly Act? An Empirical Evaluation of Agentic Capabilities in LLM Compression (ICML 2025 · similarity 0.89)
- LowRankArena: A Standardized Evaluation Platform for SVD-Based LLM Compression (NeurIPS 2026 · similarity 0.89)
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.
