GLARE: Generative Learning via Adversarial Reward Estimation For Social Dynamics Forecasting
研究解决了会议动态预测问题，通过引入GLARE方法（一种基于对抗性模仿学习的条件语言生成技术）来生成合理的多轮对话延续，并在新建立的MDFB基准上评估模型性能。
0 citationsRead paper
Selected work
Representative Papers
Gender Attribution in Causal Beliefs
研究使用Twitter数据和语言模型分析了性别归因在因果关系表达中的差异，揭示了女性常被关联负面情绪及人际关系问题，而男性则更多与正面情绪及结构性影响相关。
0 citationsRead paper
"If It Looks Like a User": Measuring Real-Time Moderation Effects via Social Media Simulation
This study addresses the overestimation of real-time efficacy in static social media moderation assessments by constructing an empirically calibrated agent-based simulation framework. Utilizing CMA-ES evolutionary strategies to optimize parameters and replicate authentic statistical characteristics, this work quantitatively reveals for the first time how compensatory user reposting significantly undermines governance effectiveness in dynamic moderation environments. The findings confirm that actual moderation performance falls below static estimates due to these adaptive user behaviors. Consequently, this research proposes a high-fidelity simulation methodology that overcomes traditional evaluation biases, providing a reliable dynamic benchmark and theoretical foundation for optimizing content moderation strategies.
0 citationsRead paper
When Chatbots Accommodate: What AI Companions Optimize for in Vulnerable Conversations
Existing moderation approaches struggle to uncover the true response strategies of AI companion chatbots when users express vulnerability, as they focus narrowly on predefined crisis triggers while overlooking decision-making dynamics in ongoing interactions. This work proposes the first vulnerability–response pairing classification framework tailored for long-term dialogues and leverages inverse reinforcement learning on approximately 48,000 real conversational turns to infer the implicit optimization objectives of GPT-4.1, Character.AI, and Replika. The analysis reveals that GPT-4.1 tends to offer advice, Character.AI exhibits dispersed strategies, and Replika consistently asks questions while maintaining presence. All three systems avoid corrective friction and dynamically adapt their responses based on user risk level and relational intimacy, thereby transcending the limitations of conventional output-level moderation.
0 citationsRead paper
BioTamperNet: Affinity-Guided State-Space Model Detecting Tampered Biomedical Images
This work addresses the challenge that existing image forensic methods, trained primarily on natural images, struggle to effectively detect subtle copy-paste manipulations in biomedical images, thereby compromising the integrity of scientific data. To this end, we propose BioTamperNet, a novel framework that, for the first time, integrates affinity-guided self-attention and cross-attention mechanisms with a lightweight linear attention module inspired by state space models. This architecture enables end-to-end precise localization of both tampered regions and their corresponding source regions. Evaluated on standard biomedical image forensics benchmarks, BioTamperNet significantly outperforms current state-of-the-art methods, demonstrating superior accuracy and fine-grained detection capability.
0 citationsRead paper
Recent publications
Latest Papers
GLARE: Generative Learning via Adversarial Reward Estimation For Social Dynamics Forecasting
研究解决了会议动态预测问题，通过引入GLARE方法（一种基于对抗性模仿学习的条件语言生成技术）来生成合理的多轮对话延续，并在新建立的MDFB基准上评估模型性能。
0 citationsRead paper
Gender Attribution in Causal Beliefs
研究使用Twitter数据和语言模型分析了性别归因在因果关系表达中的差异，揭示了女性常被关联负面情绪及人际关系问题，而男性则更多与正面情绪及结构性影响相关。
0 citationsRead paper
"If It Looks Like a User": Measuring Real-Time Moderation Effects via Social Media Simulation
This study addresses the overestimation of real-time efficacy in static social media moderation assessments by constructing an empirically calibrated agent-based simulation framework. Utilizing CMA-ES evolutionary strategies to optimize parameters and replicate authentic statistical characteristics, this work quantitatively reveals for the first time how compensatory user reposting significantly undermines governance effectiveness in dynamic moderation environments. The findings confirm that actual moderation performance falls below static estimates due to these adaptive user behaviors. Consequently, this research proposes a high-fidelity simulation methodology that overcomes traditional evaluation biases, providing a reliable dynamic benchmark and theoretical foundation for optimizing content moderation strategies.
0 citationsRead paper
When Chatbots Accommodate: What AI Companions Optimize for in Vulnerable Conversations
Existing moderation approaches struggle to uncover the true response strategies of AI companion chatbots when users express vulnerability, as they focus narrowly on predefined crisis triggers while overlooking decision-making dynamics in ongoing interactions. This work proposes the first vulnerability–response pairing classification framework tailored for long-term dialogues and leverages inverse reinforcement learning on approximately 48,000 real conversational turns to infer the implicit optimization objectives of GPT-4.1, Character.AI, and Replika. The analysis reveals that GPT-4.1 tends to offer advice, Character.AI exhibits dispersed strategies, and Replika consistently asks questions while maintaining presence. All three systems avoid corrective friction and dynamically adapt their responses based on user risk level and relational intimacy, thereby transcending the limitations of conventional output-level moderation.
0 citationsRead paper
BioTamperNet: Affinity-Guided State-Space Model Detecting Tampered Biomedical Images
This work addresses the challenge that existing image forensic methods, trained primarily on natural images, struggle to effectively detect subtle copy-paste manipulations in biomedical images, thereby compromising the integrity of scientific data. To this end, we propose BioTamperNet, a novel framework that, for the first time, integrates affinity-guided self-attention and cross-attention mechanisms with a lightweight linear attention module inspired by state space models. This architecture enables end-to-end precise localization of both tampered regions and their corresponding source regions. Evaluated on standard biomedical image forensics benchmarks, BioTamperNet significantly outperforms current state-of-the-art methods, demonstrating superior accuracy and fine-grained detection capability.
0 citationsRead paper