Research map / Learning theory and algorithms / Sample complexity
Hallucination is a Consequence of Space-Optimality: A Rate-Distortion Theorem for Membership Testing
Anxin Guo, Jingwei 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 |
|---|---|---|---|
| increase fact weight lambda_F in the training loss | from-scratch Transformer · Random strings (from scratch, 2-layer Transformer, n=15145) | hallucination rate on non-facts | Hurt |
| increase fact weight lambda_F in the training loss | from-scratch Transformer · Random strings (from scratch, 2-layer Transformer, n=15145) | information per key | Improved |
| start from a pretrained LLM prior (ISBN) instead of structure-free IDs | LoRA-tuned Qwen3.5-2B · LoRA fine-tuning (Synthetic IDs and ISBN, 12 runs) | KL overhead vs lower bound: 14-22% | No clear effect |
| fine-tune a pretrained model instead of training from scratch | Qwen3.5-2B · LoRA vs scratch | convergence speed | Improved |
| lower LoRA rank from 4 to 2 | LoRA-tuned Qwen3.5-2B · ISBN | KL per trainable parameter (bits): 0.17 -> 0.27 | Improved |
Closest papers by meaning
- Entropy Distribution as a Fingerprint for Hallucinations in Generative Models (NeurIPS 2026 · similarity 0.91)
- In-Context Sharpness as Alerts: An Inner Representation Perspective for Hallucination Mitigation (ICML 2024 · similarity 0.89)
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free Guarantees (ICML 2025 · similarity 0.89)
- Sample Transform Cost-Based Training-Free Hallucination Detector for Large Language Models (NeurIPS 2026 · similarity 0.89)
- BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation (NeurIPS 2026 · similarity 0.88)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · 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.
