collaborators

7 papers

cs.LG2026

Identified-Set Geometry of Distributional Model Extraction under Top- Censored API Access

Wenhua Nie, ZiCheng Zhu, Jianan Wu +3

Modern LLM APIs often reveal only top- logit scores and censor the remaining vocabulary. We study the per-position distribution-recovery limits of this access model. For censori…

cs.CR2026

Knowledge Poisoning Attacks on Medical Multi-Modal Retrieval-Augmented Generation

Peiru Yang, Haoran Zheng, Tong Ju +6

Retrieval-augmented generation (RAG) is a widely adopted paradigm for enhancing LLMs in medical applications by incorporating expert multimodal knowledge during generation. However…

cs.CV2026

Reflection Anchors for Propagation-Aware Visual Retention in Long-Chain Multimodal Reasoning

Xuan Gong, Hanbo Huang, Hao Zheng +4

Long chain-of-thought (CoT) reasoning improves large vision--language models, but visual information often fades during generation, limiting long-horizon multimodal reasoning. Exis…

cs.LG2026

Future Validity is the Missing Statistic: From Impossibility to -Estimation for Grammar-Faithful Speculative Decoding

Wenhua Nie, Zijie Meng, Kun Zou +5

Grammar-constrained generation is often combined with local vocabulary masking and speculative decoding, but the resulting sampling law is not the grammar-conditional distribution…

cs.LG2026

Gradient Starvation in Binary-Reward GRPO: Why Group-Mean Centering Fails and Why the Simplest Fix Works

Wenhua Nie, Jianan Wu, Junlin Liu +6

Group Relative Policy Optimization (GRPO) is a standard algorithm for reinforcement learning from verifiable rewards, but its group-mean-centered advantage can fail under binary re…

cs.LG2026

The Coupling Tax: How Shared Token Budgets Undermine Visible Chain-of-Thought Under Fixed Output Limits

Wenhua Nie, Junlin Liu, Jianan Wu +5

Chain-of-thought reasoning is often treated as a monotone way to improve language-model accuracy by letting a model think longer. We identify a countervailing effect, the coupling…