BEAGLE: Neuro-Symbolic Framework Simulates Novice Learning
A new neuro-symbolic framework named BEAGLE has been developed by researchers to emulate student learning behaviors in environments that involve open-ended problem-solving. This framework tackles the issue of competency bias prevalent in Large Language Models (LLMs), which often prioritize efficient correctness over the unpredictable, iterative challenges faced by novice learners. BEAGLE integrates Self-Regulated Learning (SRL) theory through three innovative components: a semi-Markov model that regulates the timing and transitions of cognitive and metacognitive actions, Bayesian Knowledge Tracing with intentional flaw injection to create realistic error patterns, and a neuro-symbolic architecture that merges symbolic reasoning with neural networks. This system aims to advance educational research, including the development of adaptive tutoring systems and evaluating pedagogical methods, without the need for expensive or privacy-compromising longitudinal studies. The framework is elaborated in a paper available on arXiv (2602.13280), marking a significant step forward in AI-based educational simulation.
Key facts
- BEAGLE is a neuro-symbolic framework for simulating student learning behaviors.
- It addresses competency bias in LLMs that favor efficient correctness over novice-like struggle.
- The framework incorporates Self-Regulated Learning (SRL) theory.
- Key innovations include a semi-Markov model for behavior transitions.
- Bayesian Knowledge Tracing with explicit flaw injection enforces realistic errors.
- The system combines neural networks with symbolic reasoning.
- It aims to train adaptive tutoring systems and stress-test pedagogical interventions.
- The paper is available on arXiv with identifier 2602.13280.
Entities
—