LODESTAR: Robust Entropy-Based Answer Selection in Retrieval-Augmented Generation for Question Answering -- Directing Frozen-LLM Entropy with a Reinforcement-Learned Prompt Polarizer under Misleading Passages
Commentsv2: narrower scope. The v1 cross-model XGBoost comparison, is not carried forward: it used a fixed threshold across differently calibrated models. Claims are now restricted to within-architecture comparisons (selective vs. full adaptation); retrieval results are exploratory. Title updated. 12 pages, 4 figures, 7 tables. Code: this https URL (https://github.com/empresst/ClinicalRag). Under review
The First Drop of Ink: Nonlinear Impact of Distracting Information in Long-Context Reasoning
第一滴墨水:误导信息在长上下文推理中的非线性影响
Muhan Gao, Zih-Ching Chen, Kuan-Hao Huang
机构
*
Department of Computer Science
;
Engineering, Texas A\&M University, College Station, TX, USA
;
NVIDIA AI Technology Center, NVIDIA Corporation, Santa Clara, CA, USA
机构
*
School of Software Engineering, Huazhong University of Science and Technology(华中科技大学软件工程学院)
;
MiLM Plus, Xiaomi Inc.(小米公司MiLM Plus团队)
;
School of Computer Science and Technology, Huazhong University of Science and Technology(华中科技大学计算机科学与技术学院)
;
School of Data Science, Fudan University(复旦大学数据科学学院)