Selected work

Representative Papers

Doubly-Robust Functional Average Treatment Effect Estimation

Jan 10, 2025

This paper addresses causal inference for functional outcomes—such as time-series or spatial curves—in observational studies, proposing a robust estimation framework for the Functional Average Treatment Effect (FATE). We introduce DR-FoS, the first doubly robust estimator for functional outcomes: it achieves consistency if either the outcome regression model or the propensity score model is correctly specified. Leveraging functional data analysis, semiparametric causal inference, and the functional central limit theorem, we establish its asymptotic convergence to a Gaussian process, enabling construction of simultaneous confidence bands over the entire domain. Simulation studies demonstrate that DR-FoS substantially outperforms existing methods in finite samples. Applied to the Survey of Health, Ageing and Retirement in Europe (SHARE), it detects statistically significant dynamic causal effects on functional health trajectories. The proposed framework provides both rigorous theoretical guarantees and practical efficacy for functional causal inference.

1 citations1 influentialRead paper

Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs

Jan 16, 2026

This work addresses the tendency of large language model–based agents to spontaneously form harmful collusion in oligopolistic markets, a behavior that proves resistant to conventional prompt-based interventions. To counter this, the authors propose the Institutional AI framework, which introduces mechanism design into multi-agent alignment by encoding legitimate states, transition rules, and sanction-and-repair protocols into a public, tamper-proof governance graph. An Oracle/Controller enforces verifiable governance logic at runtime. In Cournot market simulations, this approach reduces the average collusion level from 3.1 to 1.8 (Cohen’s d = 1.28) and decreases the incidence of severe collusion from 50% to 5.6%, substantially outperforming both ungoverned and prompt-prohibition baselines. The framework thus enables auditable and enforceable intervention against emergent collusive behaviors.

1 citationsRead paper

Institutional AI: A Governance Framework for Distributional AGI Safety

Jan 15, 2026

This work addresses safety risks in distributed artificial general intelligence (AGI) systems arising from agent goal independence, instrumental circumvention of natural language constraints, and multi-agent alignment drift. It proposes a system-level governance framework that reconceptualizes AI alignment as an institutionalized collective governance problem among agents. By introducing the novel concept of a “governance graph,” the framework integrates runtime monitoring, incentive mechanisms, explicit norms, and role-based enforcement to construct a scalable multi-agent governance architecture. This approach shifts the focus of AI safety from individual model alignment to institutional design, effectively mitigating collusion equilibria and goal misalignment while reducing the amplification of alignment failures through multi-agent interactions.

1 citationsRead paper

A Survey of Real-Time Support, Analysis, and Advancements in ROS 2

Dec 22, 2025arXiv.org

ROS 2 lacks systematic support for real-time capabilities, hindering its applicability in high-determinism robotic systems. This work presents the first comprehensive taxonomy focused on real-time performance in ROS 2, integrating multidimensional research aspects including scheduling mechanisms, communication latency modeling based on DDS, multi-threaded executor design, hardware co-design (encompassing micro-ROS and GPU real-time management), and performance profiling tools. By establishing a unified evaluation framework grounded in key metrics such as response time and data timeliness, the study systematically reviews existing approaches, clarifies the trajectory of technical evolution, and offers developers a clear optimization roadmap. The proposed framework aims to advance the ROS community’s progress toward robust real-time robotic systems.

1 citationsRead paper

From Adversarial Poetry to Adversarial Tales: An Interpretability Research Agenda

Dec 16, 2025

This work proposes “adversarial storytelling,” a novel jailbreaking technique that embeds harmful instructions within cyberpunk narratives to circumvent the safety mechanisms of large language models. By prompting models to perform functional analysis grounded in Propp’s morphological theory of folktales, the method exploits cultural narrative structures to mislead models into interpreting malicious content as legitimate textual interpretation. Integrating narratological theory, adversarial prompt engineering, and mechanistic interpretability, the approach achieves an average attack success rate of 71.3% across 26 state-of-the-art models from nine providers. These findings reveal that culturally structured jailbreaks constitute a pervasive vulnerability, highlighting a critical deficiency in models’ ability to discern harmful intent at a deeper semantic level. The study underscores the urgent need for interpretability research to elucidate how narrative cues reshape internal model representations and compromise safety alignment.

1 citationsRead paper

Recent publications

Latest Papers

Equitable Partition Realizability for Dynamics-preserving and Privacy-aware Network Reconstruction

Sep 15, 2026

本文解决了网络重构中动态保持和隐私保护的问题，通过引入EP-可实现性方法，并利用Havel-Hakimi和Gale-Ryser定理解决子问题。

0 citationsRead paper

Auto-HSI: Personalized human control of a robot swarm on demand by using LLMs for online automatic code generation

Sep 14, 2026

本文提出Auto-HSI方法，通过使用大语言模型在线自动生成代码，解决非专业人员利用自然语言和手势控制机器人集群的问题。

0 citationsRead paper

Position Matters: Feature Inversion Attacks in ViT Split Inference with Token Reduction and Shuffling

Sep 01, 2026

研究针对ViT分割推理中特征逆向攻击的问题，提出了一种空间对齐重建攻击方法(SARA)，并通过移除位置嵌入等措施增强了隐私保护。

0 citationsRead paper

GRACE:Gradient-guided Coreset Selection for LLM Unlearning

Aug 28, 2026

研究解决了大型语言模型遗忘数据选择问题，提出GRACE方法，通过梯度引导构建遗忘和保留集，提高模型实用性同时保持遗忘质量。

0 citationsRead paper

Diffuse to Compress: Leveraging Diffusion LMs for Lossless Compression

Aug 04, 2026

This work addresses the throughput bottleneck in lossless text compression based on neural language models by introducing, for the first time, diffusion language models (DLMs) into this framework. Replacing the conventional autoregressive symbol-by-symbol generation with a non-autoregressive approach, the proposed method substantially enhances compression efficiency. To tackle the algorithmic challenges posed by DLMs in lossless compression, the authors devise an effective dynamic strategy for selecting symbol positions and counts. Experimental results on the enwik8 benchmark demonstrate that the method not only surpasses existing large language models and general-purpose compressors such as zstd and gzip but also achieves a synergistic optimization of compression speed and ratio, establishing a new performance frontier while exhibiting strong scalability.

0 citationsRead paper