Cognitive Structure Generation via Diffusion Models with Policy Optimization

ICLR 2026 Conference Submission15303 Authors

Keywords: Cognitive structure

TL;DR: We propose Cognitive Structure Generation (CSG), a generative framework that learns cognitive structures and decouples representation from downstream prediction.

Abstract: Cognitive structure (CS), a student's construction of concepts and inter-concept relations, has long been recognized as a foundational notion in educational psychology, yet remains largely unassessable in practice. Existing approaches such as knowledge tracing (KT) and cognitive diagnosis (CD) simplify and indirectly approximate CS, but they intertwine representation learning with prediction objectives, limiting generalization, interpretability, and reuse across tasks. To address this gap, we propose Cognitive Structure Generation (CSG), a task-agnostic framework that explicitly models CS through generative modeling. Based on educational theories, CSG first pretrains a Cognitive Structure Diffusion Probabilistic Model (CSDPM) and then applies reinforcement learning with SOLO-based hierarchical rewards to align generation with genuine cognitive development. By decoupling cognitive structure representation from downstream prediction, CSG produces interpretable and transferable cognitive structures that can be seamlessly integrated into diverse student modeling tasks. Experiments on four real-world datasets show that CSG yields more comprehensive representations, substantially improving performance while offering enhanced interpretability and modularity.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 15303