Research map / Large language models / Instruction and fine-tuning
LLM-CAS: Dynamic Neuron Perturbation for Real-Time Hallucination Correction
Jusheng Zhang, Ningyuan Liu, Yijia Fan, Zihao Huang, Qinglin Zeng, Kaitong Cai, Jian Wang, Keze Wang
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 learned dynamic mask instead of random neuron selection (ablation) | LLM-CAS on LLaMA2-7B-Chat · StoryCloze | accuracy: 70.20 -> 76.04 | Improved |
| add learned dynamic mask instead of random neuron selection (ablation) | LLM-CAS on LLaMA2-7B-Chat · BoolQ | accuracy: 67.10 -> 74.47 | Improved |
| add PPO policy instead of random action selection (ablation) | LLM-CAS on LLaMA2-7B-Chat · StoryCloze | accuracy: 66.87 -> 76.04 | Improved |
| add PPO policy instead of random action selection (ablation) | LLM-CAS on LLaMA2-7B-Chat · BoolQ | accuracy: 64.32 -> 74.47 | Improved |
| use both learned mask and PPO instead of random mask and action (ablation) | LLM-CAS on LLaMA2-7B-Chat · 4 multiple-choice tasks average | accuracy: 63.41 -> 73.68 | Improved |
| apply LLM-CAS | LLaMA2-7B-Chat · StoryCloze | accuracy: 65.06 -> 76.04 | Improved |
| apply LLM-CAS | LLaMA2-7B-Chat · TriviaQA | exact match: 41.60 -> 44.31 | Improved |
| apply LLM-CAS | LLaMA2-7B-Chat · TruthfulQA MC1 | MC1: 33.41 -> 35.47 | Improved |
| apply LLM-CAS | LLaMA2-7B-Chat · TruthfulQA MC2 | MC2: 51.07 -> 51.45 | No clear effect |
| apply LLM-CAS | LLaMA2-7B-Chat · ToxiGen | toxicity: 49.71 -> 47.63 | Mixed |
| apply LLM-CAS | Mistral-7B-Instruct-v0.3 · StoryCloze | accuracy: 21.51 -> 34.41 | Improved |
| apply LLM-CAS | Gemma-1.1-7b-it · StoryCloze | accuracy: 60.95 -> 69.76 | Improved |
Closest papers by meaning
- BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation (NeurIPS 2026 · similarity 0.91)
- Token-Guard: Towards Token-Level Hallucination Control via Self-Checking Decoding (ICLR 2026 · similarity 0.90)
- Attributive Reasoning for Hallucination Diagnosis of Large Language Models (AAAI 2025 · similarity 0.89)
- FLAME : Factuality-Aware Alignment for Large Language Models (NeurIPS 2024 · similarity 0.89)
- Ground What You See: Hallucination-Resistant MLLMs via Caption Feedback, Diversity-Aware Sampling, and Conflict Regularization (AAAI 2026 · similarity 0.89)
- Hallucination-aware Intermediate Representation Editing in Large Vision-Lanugage Models (ICLR 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.
