The Academic Salon of the Department of Psychology and Cognitive Sciences is held every Thursday afternoon. Welcome to all students and faculty members from every department!
Time: April 24 (Thursday) afternoon 15:00
Location: Room 1110, 11th floor, Lv Dalong Building
Department of Psychology and Cognitive Sciences STELLA's Research Group Academic Salon
Report One
Know Little, Explore More? The Role of
Self-Assessment and Perceived
Knowledge Gaps in
Exploration-Exploitation Decisions
Presenter: Yijin Fang
Content:
Exploration-exploitation decisions are prevalent in people's daily life and learning, requiring individuals to balance between seeking new information and relying on existing information. Such decisions are often influenced by individuals' knowledge states. However, how individuals attend to and identify their own knowledge states and influence exploration decisions still lacks deep understanding. On one hand, people may not always attend to and monitor their knowledge foundation, especially when lacking clear learning objectives; on the other hand, people's perception of their own knowledge may also deviate from reality. This study explores how individuals' attention to and perception of their own knowledge states affect their exploration-exploitation decisions. The study found that regardless of how much knowledge was actually mastered, individuals who were induced to have "insufficient" knowledge perception explored more. Additionally, individuals' maintaining awareness of their knowledge states increased exploration, a trend that was more pronounced in individuals who believed their knowledge states had more room for growth. This study emphasizes the key role of metacognitive processes of knowledge states in shaping exploration-exploitation behavior and attempts to understand changes in exploration-exploitation decisions from the perspective of goals and motivation.
Report Two
Reinforcement Learning-Based Rule Abstraction Model
RuleFinder: Human-Machine Cognitive-Algorithmic
Decoupling Mechanism Under Data Diversity Constraints
Presenter: Qiuchen Ma
Content:
This study proposes RuleFinder, a reinforcement learning-based rule abstraction model that effectively improves generalization ability to new instances and isomorphic new rules through feature similarity calculation of contrastive modules and dynamic rule update mechanisms. Experiments show that RuleFinder performs robustly in strict generalization tests containing new instances and isomorphic new rules, with accuracy far exceeding baseline models. Through systematic comparison of human participants and RuleFinder in 3-token combination tasks, this study reveals: 1) When training rules decreased from 6 to 2, RuleFinder's generalization accuracy dropped sharply, while human participants maintained stable performance, confirming AI performance's high dependence on training data diversity; 2) Human rule abstraction is susceptible to interference from specific attributes (symmetry), while RuleFinder shows strong robustness to such interference. The study shows that RuleFinder successfully achieves transferable rule representation learning through dynamic rule update mechanisms driven by contrastive feature similarity. This study provides a reproducible experimental paradigm and verifiable computational framework for understanding the heterogeneity of artificial intelligence and human rule learning mechanisms.
Report Three
Control and Agency: The Influence of Parental
Parenting Beliefs on Children's Agency
Presenter: Weiwei Fan
Content:
Agency, as a core concept in children's cognitive development, presents a paradox of goals and means in its developmental pathway within parenting cultures that emphasize control: Chinese parents' "control" belief emphasizes cultivating children's self-discipline and active learning ability through strict control. While this goal aligns with core characteristics of agency (such as initiative and self-regulation), its emphasis on control may instead limit agency development by undermining autonomy. This study aims to explore the coexistence mechanism of parental "control" and agency beliefs and their impact on children's agency. Study 1, through developing a dual-dimensional questionnaire of "control" and agency beliefs, found a U-shaped relationship between the two: moderate "control" belief corresponded to the lowest agency belief; while higher or lower levels of "control" belief corresponded to higher agency beliefs, revealing a dialectical regulatory mechanism between parental "control" and agency beliefs. Study 2 focused on family decision-making contexts to examine the impact of parental "control" and agency beliefs on children's agency. Results found that in "learning" decision contexts, control-oriented "control" belief negatively predicted children's agency perception, while agency belief positively predicted it; in low agency contexts, children's agency perception significantly predicted their interpretation of maternal controlling decisions and proactive coping strategies, indicating that even under strong external constraints, children can still activate proactive behaviors through agency perception. The study reveals: parental belief transformation, from emphasizing control to supporting children's agency, helps promote children's agency development in learning contexts.
