Causal Modeling of Adverse Pregnancy Outcomes via Adaptive LLM Proposals
该研究通过结合大型语言模型和实证数据评分的神经符号框架，生成关于不良妊娠结局的合理因果假设，以解决数据稀缺和领域知识不完整的问题。
0 citationsRead paper
Selected work
Representative Papers
One Adaptive Trailing Head Can Outperform Many Oblivious Trailing Heads
This study investigates the advantage of adaptive trailing heads over arbitrarily many non-adaptive trailing heads in sequence predictability within the framework of finite-state strong dimension. By integrating multi-head finite-state automata, finite-state dimension theory, and information-theoretic analysis, the authors construct a binary sequence for which the strong dimension under an adaptive two-head model is at least 0.3 lower than that under any non-adaptive multi-head model. This result demonstrates—by a substantial and consistent margin—that even a single adaptive trailing head can outperform any number of non-adaptive heads employing fixed strategies. The finding strengthens and extends existing dimension separation results, highlighting the fundamental superiority of adaptive strategies in finite-state prediction.
0 citationsRead paper
ScamPilot: Simulating Conversations with LLMs to Protect Against Online Scams
This study addresses the growing challenge of evolving online scams, which outpace existing automated defense systems in equipping users to recognize novel fraud tactics. To bridge this gap, the authors propose a conversational anti-fraud training framework powered by large language models, featuring two interacting agents—one simulating a scammer and the other a potential victim—to dynamically recreate realistic scam scenarios. The approach integrates real-time user intervention with multiple-choice prompts, encouraging participants to provide actionable advice that reinforces fraud awareness. In a controlled experiment involving 150 participants, the method significantly improved scam identification accuracy by 8%, response effectiveness by 9%, and self-efficacy by 19%. Notably, users predominantly offered action-oriented recommendations without compromising trust in legitimate interactions.
0 citationsRead paper
Multihead Finite-State Compression
Characterizing the optimal lossless compression ratio achievable by finite-state compressors with multiple forward-only reading heads scanning infinite symbol sequences, and establishing its precise relationship to algorithmic dimension. Method: The authors extend finite-state compression theory by introducing a multi-head finite-state lossless compression model and defining the $h$-head finite-state prediction dimension—a new dimension notion based on finite-state predictability using $h$ synchronized reading heads. Contribution/Results: They prove that, for each fixed $h$, the infimum of achievable compression ratios equals the $h$-head finite-state prediction dimension; moreover, the supremum of these infima over all $h$ yields the multi-head finite-state dimension of the sequence. This work establishes, for the first time, an exact equivalence between multi-head finite-state compressibility and prediction-based dimension, providing a novel theoretical bridge between algorithmic information theory and data compression.
0 citationsRead paper
Promoting Online Safety by Simulating Unsafe Conversations with LLMs
Public online security awareness remains low, and users struggle to recognize deceptive conversational patterns in social engineering scams. Method: This study proposes an interactive, large language model (LLM)-driven security education framework. It employs a dual-LLM adversarial simulation architecture to autonomously generate high-fidelity, diverse scam dialogues; integrates principles from learning science—including just-in-time feedback and guided reflection—to help users identify linguistic cues, assess risks, and practice defensive responses. Contribution/Results: Empirical evaluation demonstrates significant improvements in both scam detection accuracy and willingness to enact protective behaviors. This work constitutes the first systematic validation of LLM-powered, scenario-based simulation for cybersecurity literacy education—establishing its efficacy, scalability, and pedagogical viability. It introduces a novel, AI-augmented paradigm for security education grounded in authentic, adaptive interaction.
0 citationsRead paper
Recent publications
Latest Papers
Causal Modeling of Adverse Pregnancy Outcomes via Adaptive LLM Proposals
该研究通过结合大型语言模型和实证数据评分的神经符号框架，生成关于不良妊娠结局的合理因果假设，以解决数据稀缺和领域知识不完整的问题。
0 citationsRead paper
One Adaptive Trailing Head Can Outperform Many Oblivious Trailing Heads
This study investigates the advantage of adaptive trailing heads over arbitrarily many non-adaptive trailing heads in sequence predictability within the framework of finite-state strong dimension. By integrating multi-head finite-state automata, finite-state dimension theory, and information-theoretic analysis, the authors construct a binary sequence for which the strong dimension under an adaptive two-head model is at least 0.3 lower than that under any non-adaptive multi-head model. This result demonstrates—by a substantial and consistent margin—that even a single adaptive trailing head can outperform any number of non-adaptive heads employing fixed strategies. The finding strengthens and extends existing dimension separation results, highlighting the fundamental superiority of adaptive strategies in finite-state prediction.
0 citationsRead paper
ScamPilot: Simulating Conversations with LLMs to Protect Against Online Scams
This study addresses the growing challenge of evolving online scams, which outpace existing automated defense systems in equipping users to recognize novel fraud tactics. To bridge this gap, the authors propose a conversational anti-fraud training framework powered by large language models, featuring two interacting agents—one simulating a scammer and the other a potential victim—to dynamically recreate realistic scam scenarios. The approach integrates real-time user intervention with multiple-choice prompts, encouraging participants to provide actionable advice that reinforces fraud awareness. In a controlled experiment involving 150 participants, the method significantly improved scam identification accuracy by 8%, response effectiveness by 9%, and self-efficacy by 19%. Notably, users predominantly offered action-oriented recommendations without compromising trust in legitimate interactions.
0 citationsRead paper
Multihead Finite-State Compression
Characterizing the optimal lossless compression ratio achievable by finite-state compressors with multiple forward-only reading heads scanning infinite symbol sequences, and establishing its precise relationship to algorithmic dimension. Method: The authors extend finite-state compression theory by introducing a multi-head finite-state lossless compression model and defining the $h$-head finite-state prediction dimension—a new dimension notion based on finite-state predictability using $h$ synchronized reading heads. Contribution/Results: They prove that, for each fixed $h$, the infimum of achievable compression ratios equals the $h$-head finite-state prediction dimension; moreover, the supremum of these infima over all $h$ yields the multi-head finite-state dimension of the sequence. This work establishes, for the first time, an exact equivalence between multi-head finite-state compressibility and prediction-based dimension, providing a novel theoretical bridge between algorithmic information theory and data compression.
0 citationsRead paper
Promoting Online Safety by Simulating Unsafe Conversations with LLMs
Public online security awareness remains low, and users struggle to recognize deceptive conversational patterns in social engineering scams. Method: This study proposes an interactive, large language model (LLM)-driven security education framework. It employs a dual-LLM adversarial simulation architecture to autonomously generate high-fidelity, diverse scam dialogues; integrates principles from learning science—including just-in-time feedback and guided reflection—to help users identify linguistic cues, assess risks, and practice defensive responses. Contribution/Results: Empirical evaluation demonstrates significant improvements in both scam detection accuracy and willingness to enact protective behaviors. This work constitutes the first systematic validation of LLM-powered, scenario-based simulation for cybersecurity literacy education—establishing its efficacy, scalability, and pedagogical viability. It introduces a novel, AI-augmented paradigm for security education grounded in authentic, adaptive interaction.
0 citationsRead paper