Scholarlink

<p>This study explores expert–novice differences in anticipation under uncertainty by combining partially observable Markov decision process (POMDP) modeling with machine learning classification. Forty-eight participants (24 experts, 24 novices) completed a basketball pass/shot anticipation task. Through POMDP modeling, two cognitive parameters—sensory precision (SP) and prior belief (pB)—were extracted to capture internal decision processes. Results showed that experts fit the POMDP model more closely, requiring more iterations for parameter convergence and achieving higher pseudo <i>R</i><sup><i>2</i></sup> values than novices. Experts demonstrated significantly higher SP, indicating superior ability to filter key cues under noisy conditions. Their pB values remained closer to neutral, suggesting flexible reliance on prior knowledge. In contrast, novices exhibited more biased priors and a lower, more dispersed SP. Machine learning analyses revealed that SP and pB jointly formed distinct clusters for experts and novices in a two-dimensional parameter space, with classification accuracies exceeding 90% across multiple methods. These findings indicate that expertise entails both enhanced perceptual precision and adaptive prior calibration, reflecting deeper cognitive reorganization rather than simple skill increments. Our dual-parameter approach offers a model-based perspective on expert cognition and may inform future research on the multifaceted nature of expertise.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Beyond performance: A POMDP-based machine learning framework for expert cognition

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • Hao He,
  • Yucheng Duan

摘要

This study explores expert–novice differences in anticipation under uncertainty by combining partially observable Markov decision process (POMDP) modeling with machine learning classification. Forty-eight participants (24 experts, 24 novices) completed a basketball pass/shot anticipation task. Through POMDP modeling, two cognitive parameters—sensory precision (SP) and prior belief (pB)—were extracted to capture internal decision processes. Results showed that experts fit the POMDP model more closely, requiring more iterations for parameter convergence and achieving higher pseudo R2 values than novices. Experts demonstrated significantly higher SP, indicating superior ability to filter key cues under noisy conditions. Their pB values remained closer to neutral, suggesting flexible reliance on prior knowledge. In contrast, novices exhibited more biased priors and a lower, more dispersed SP. Machine learning analyses revealed that SP and pB jointly formed distinct clusters for experts and novices in a two-dimensional parameter space, with classification accuracies exceeding 90% across multiple methods. These findings indicate that expertise entails both enhanced perceptual precision and adaptive prior calibration, reflecting deeper cognitive reorganization rather than simple skill increments. Our dual-parameter approach offers a model-based perspective on expert cognition and may inform future research on the multifaceted nature of expertise.