Can Knowledge Transfer Parameters Be Learned? LePoKet for Efficient Robotic Vision
该研究通过LePoKet框架解决机器人视觉中高效感知的问题，利用可学习参数优化知识转移过程，无需辅助蒸馏损失或温度缩放。
0 citationsRead paper
Selected work
Representative Papers
A race to belief: How Evidence Accumulation shapes trust in AI and Human informants
This study investigates the cognitive mechanisms underlying differential trust in AI versus human information providers across factual and social contexts. Using a Bayesian hierarchical sequential sampling model (HSSM), we analyzed trust decisions from participants across 30 distinct scenarios. Results indicate that trust preferences are primarily driven by the evidence accumulation rate (drift rate), rather than prior biases or decision caution. A key contribution is the identification of domain-specific vigilance mechanisms: in factual contexts, negative drift rates accelerate erosion of trust in AI, whereas in social contexts, positive drift rates strengthen trust in humans; critically, drift rates correlate significantly with real-time confidence judgments. These findings establish a novel, dynamic evidence-accumulation framework for understanding the cognitive roots of AI trust fragility, moving beyond static attributional accounts to reveal how context-dependent accumulation dynamics shape trust formation in real time.
0 citationsRead paper
Trust in AI emerges from distrust in humans: A machine learning study on decision-making guidance
This study investigates the formation mechanism of human trust in AI agents—particularly large language models—proposing the “transferrable trust” theory: when users distrust human agents, they strategically shift reliance toward AI perceived as more neutral, reliable, or competent, resulting in compensatory dependence. Method: Integrating sociodemographic and prior-trust variables, the study employs K-Modes/K-Means clustering to identify behavioral patterns and builds an interpretable XGBoost-SHAP model to predict AI adoption propensity. Contribution/Results: Low trust in humans, limited technology experience, and high socioeconomic status emerge as key predictors of strong AI trust. AI adoption reaches 28.29% in factual decision-making, whereas humans retain superiority in social/moral contexts. The model achieves a mean accuracy of 0.863. This work transcends conventional technology acceptance frameworks by systematically uncovering how social relational dynamics and cognitive distrust drive trust migration toward AI.
0 citationsRead paper
Recent publications
Latest Papers
Can Knowledge Transfer Parameters Be Learned? LePoKet for Efficient Robotic Vision
该研究通过LePoKet框架解决机器人视觉中高效感知的问题，利用可学习参数优化知识转移过程，无需辅助蒸馏损失或温度缩放。
0 citationsRead paper
A race to belief: How Evidence Accumulation shapes trust in AI and Human informants
This study investigates the cognitive mechanisms underlying differential trust in AI versus human information providers across factual and social contexts. Using a Bayesian hierarchical sequential sampling model (HSSM), we analyzed trust decisions from participants across 30 distinct scenarios. Results indicate that trust preferences are primarily driven by the evidence accumulation rate (drift rate), rather than prior biases or decision caution. A key contribution is the identification of domain-specific vigilance mechanisms: in factual contexts, negative drift rates accelerate erosion of trust in AI, whereas in social contexts, positive drift rates strengthen trust in humans; critically, drift rates correlate significantly with real-time confidence judgments. These findings establish a novel, dynamic evidence-accumulation framework for understanding the cognitive roots of AI trust fragility, moving beyond static attributional accounts to reveal how context-dependent accumulation dynamics shape trust formation in real time.
0 citationsRead paper
Trust in AI emerges from distrust in humans: A machine learning study on decision-making guidance
This study investigates the formation mechanism of human trust in AI agents—particularly large language models—proposing the “transferrable trust” theory: when users distrust human agents, they strategically shift reliance toward AI perceived as more neutral, reliable, or competent, resulting in compensatory dependence. Method: Integrating sociodemographic and prior-trust variables, the study employs K-Modes/K-Means clustering to identify behavioral patterns and builds an interpretable XGBoost-SHAP model to predict AI adoption propensity. Contribution/Results: Low trust in humans, limited technology experience, and high socioeconomic status emerge as key predictors of strong AI trust. AI adoption reaches 28.29% in factual decision-making, whereas humans retain superiority in social/moral contexts. The model achieves a mean accuracy of 0.863. This work transcends conventional technology acceptance frameworks by systematically uncovering how social relational dynamics and cognitive distrust drive trust migration toward AI.
0 citationsRead paper