Efficient Parameter Estimation of Cognitive Models for Real-Time Performance Monitoring and Adaptive Interfaces
Efficient Parameter Estimation of Cognitive Models for Real-Time Performance Monitoring and Adaptive Interfaces
复制标题
用于实时性能监控和自适应界面的认知模型的有效参数估计
DOI：
--
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
F. Ritter
中科院分区：
文献类型：
--
作者：
Christopher R. Fisher;Matthew M. Walsh;L. Blaha;G. Gunzelmann;In D. Reitter;F. Ritter
实时监控提供了一个机会，可以检查认知的时间动态，预测未来行为并实施旨在减轻性能下降的自适应界面。但是，实时监视会带来实际的挑战，因为当前参数估计方法非常缓慢，并且实时监视需要在新数据到达时反复估算参数。我们开发了一种实时参数估计方法，该方法涉及将预计的预测存储在分布式阵列中，并将其作为大型查找表。我们称此方法为预计的分布式查找表（PDLT）。我们将PDLT应用于精神病警戒测试的ACT-R模型。 PDLT估计模型参数仅1秒钟以上，其精度与慢速单纯形方法的准确性相当。我们讨论了降低参数估计的波动性以及将PDLT方法扩展到更复杂的模型和任务的潜力的方法。
Real-time monitoring provides an opportunity to examine the temporal dynamics of cognition, predict future behavior, and implement adaptive interfaces designed to mitigate declining performance. However, real-time monitoring poses a practical challenge because current parameter estimation methods are prohibitively slow, and real-time monitoring requires parameters to be estimated repeatedly as new data arrive. We developed a real-time parameter estimation method that involves storing pre-computed predictions in a distributed array and us-ing it as a large look-up table. We term this method the Pre-computed Distributed Look-up Table (PDLT). We applied the PDLT to an ACT-R model of the psychomotor vigilance test. PDLT estimates model parameters in just over 1 second with accuracy comparable to that of a much slower simplex method. We discuss methods for reducing the volatility of parameter estimates and the potential to scale up the PDLT method to more complex models and tasks.