Selected work

Representative Papers

Invisible Ink, Visible Lies: How Production Watermarking Causes LLMs to Hallucinate

Oct 04, 2026

This study addresses a critical side effect of production-grade text watermarking for large language models: while preserving detectability, watermarking induces factual hallucinations, causing models to generate incorrect content that disregards evidence. This work systematically quantifies this phenomenon for the first time, revealing that it stems from a dual failure mechanism involving token perturbation and attention drift. To mitigate this issue, we propose a plug-and-play correction strategy based on token reweighting and attention optimization, which is compatible with mainstream watermarking algorithms such as RAG and KGW. Experimental results demonstrate that the proposed method reduces factual errors by approximately 90% while maintaining both watermark detection rates and text fluency. Furthermore, this work establishes factuality as a core evaluation metric for text watermarking systems.

0 citationsRead paper

SteerProbe: Learning to Bypass Safety Steering in Vision-Language Models

Sep 30, 2026

This study addresses the vulnerability of existing safety-aligned defenses in vision-language models to intent-preserving input reconstruction attacks. We propose SteerProbe, a black-box attack framework that leverages calibration learning to perform activation-steered local reconstruction of multimodal inputs. By establishing sufficient conditions for traversing safety boundaries, SteerProbe enables efficient, automated output-side attacks. Experimental results demonstrate that our method significantly increases the average harmfulness rate from 7.43% to 18.36%, exposing critical robustness deficiencies in current defense mechanisms. This work provides a novel perspective for evaluating and enhancing the safety of multimodal large language models.

0 citationsRead paper

HeadWiseKV: Budgeted Per-Head Cache Residency for Hybrid Long-Context Language Models

Sep 01, 2026

为解决长上下文推理中GPU内存消耗大、生成吞吐量低的问题，提出HeadWiseKV框架，通过压缩全局KV缓存并分配静态多级历史窗口来优化缓存需求。

0 citationsRead paper

Heterogeneous Effects of Green Finance on Urban Decarbonization: Evidence from 285 Cities in China

Jun 05, 2026

This study addresses the ambiguous empirical effectiveness and underlying mechanisms of green finance in facilitating urban low-carbon transitions, particularly the lack of systematic analysis on heterogeneity and regional disparities. Leveraging a dataset covering 285 Chinese cities, the authors integrate econometric and machine learning approaches to evaluate the impact of green finance on carbon intensity and its transmission channels through energy structure optimization and industrial upgrading. The findings reveal that the carbon-mitigation effects of green finance are more pronounced in cities characterized by weaker technological capabilities, higher industrial dependence, and greater coal consumption shares. Green bonds and green investments exhibit the strongest efficacy and generate significant spatial spillovers, with tier-four and tier-five cities deriving the greatest benefits. These results provide empirical support for designing a multi-tiered and differentiated green finance system.

0 citationsRead paper

Recent publications

Latest Papers

Invisible Ink, Visible Lies: How Production Watermarking Causes LLMs to Hallucinate

Oct 04, 2026

This study addresses a critical side effect of production-grade text watermarking for large language models: while preserving detectability, watermarking induces factual hallucinations, causing models to generate incorrect content that disregards evidence. This work systematically quantifies this phenomenon for the first time, revealing that it stems from a dual failure mechanism involving token perturbation and attention drift. To mitigate this issue, we propose a plug-and-play correction strategy based on token reweighting and attention optimization, which is compatible with mainstream watermarking algorithms such as RAG and KGW. Experimental results demonstrate that the proposed method reduces factual errors by approximately 90% while maintaining both watermark detection rates and text fluency. Furthermore, this work establishes factuality as a core evaluation metric for text watermarking systems.

0 citationsRead paper

SteerProbe: Learning to Bypass Safety Steering in Vision-Language Models

Sep 30, 2026

This study addresses the vulnerability of existing safety-aligned defenses in vision-language models to intent-preserving input reconstruction attacks. We propose SteerProbe, a black-box attack framework that leverages calibration learning to perform activation-steered local reconstruction of multimodal inputs. By establishing sufficient conditions for traversing safety boundaries, SteerProbe enables efficient, automated output-side attacks. Experimental results demonstrate that our method significantly increases the average harmfulness rate from 7.43% to 18.36%, exposing critical robustness deficiencies in current defense mechanisms. This work provides a novel perspective for evaluating and enhancing the safety of multimodal large language models.

0 citationsRead paper

HeadWiseKV: Budgeted Per-Head Cache Residency for Hybrid Long-Context Language Models

Sep 01, 2026

为解决长上下文推理中GPU内存消耗大、生成吞吐量低的问题，提出HeadWiseKV框架，通过压缩全局KV缓存并分配静态多级历史窗口来优化缓存需求。

0 citationsRead paper

Heterogeneous Effects of Green Finance on Urban Decarbonization: Evidence from 285 Cities in China

Jun 05, 2026

This study addresses the ambiguous empirical effectiveness and underlying mechanisms of green finance in facilitating urban low-carbon transitions, particularly the lack of systematic analysis on heterogeneity and regional disparities. Leveraging a dataset covering 285 Chinese cities, the authors integrate econometric and machine learning approaches to evaluate the impact of green finance on carbon intensity and its transmission channels through energy structure optimization and industrial upgrading. The findings reveal that the carbon-mitigation effects of green finance are more pronounced in cities characterized by weaker technological capabilities, higher industrial dependence, and greater coal consumption shares. Green bonds and green investments exhibit the strongest efficacy and generate significant spatial spillovers, with tier-four and tier-five cities deriving the greatest benefits. These results provide empirical support for designing a multi-tiered and differentiated green finance system.

0 citationsRead paper