Scholarlink

<p>As a stable and reliable paradigm, P300-based brain-computer interface (P300-BCI) is expected to play an important role in efforts to replace, restore, enhance, supplement, or improve the natural output of the brain. However, the costly calibration of P300-BCI limits its development. The calibration-free approaches for P300-BCI have become a research focus in the field. In this work, we forwarded our previous study, transferred P300 linear upper confidence bound (TPLUCB), to propose an adaptive ensemble P300-BCI classifier (AEP). This renovation mainly includes a simplified calculation method and a dynamical update strategy. The competitive calculation model in TPLUCB was simplified as a linear calculation model. Based on this, a dynamical update strategy was proposed to facilitate the growth of target domain model and optimize the weights, by which the source domain models and the target domain model are combined as a P300-BCI classifier, <i>i.e.</i> AEP. We conducted the performance evaluation by observing the classifier’s dynamical development and overall performance. The comparison in the two aspects between AEP and TPLUCB demonstrates AEP’s clear advantage over TPLUCB. Without prior calibration, AEP achieved an average ITR exceeding 40 bit/min on electroencephalogram (EEG) data of 20 subjects. This work has provided a better calibration-free approach for P300-BCI and is an important step towards promoting the research on calibration-free BCIs.</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

AEP: An adaptive ensemble P300-BCI classifier based on user-feedback and knowledge-transfer

  • 期刊论文

  • 发表日期：

    2025年5月9日

查看全文
  • Zhihua Huang,
  • Qingzhi Chen,
  • Xuewei Chen,
  • Wenming Zheng,
  • Zhixiong Lin,
  • Tian-jian Luo

摘要

As a stable and reliable paradigm, P300-based brain-computer interface (P300-BCI) is expected to play an important role in efforts to replace, restore, enhance, supplement, or improve the natural output of the brain. However, the costly calibration of P300-BCI limits its development. The calibration-free approaches for P300-BCI have become a research focus in the field. In this work, we forwarded our previous study, transferred P300 linear upper confidence bound (TPLUCB), to propose an adaptive ensemble P300-BCI classifier (AEP). This renovation mainly includes a simplified calculation method and a dynamical update strategy. The competitive calculation model in TPLUCB was simplified as a linear calculation model. Based on this, a dynamical update strategy was proposed to facilitate the growth of target domain model and optimize the weights, by which the source domain models and the target domain model are combined as a P300-BCI classifier, i.e. AEP. We conducted the performance evaluation by observing the classifier’s dynamical development and overall performance. The comparison in the two aspects between AEP and TPLUCB demonstrates AEP’s clear advantage over TPLUCB. Without prior calibration, AEP achieved an average ITR exceeding 40 bit/min on electroencephalogram (EEG) data of 20 subjects. This work has provided a better calibration-free approach for P300-BCI and is an important step towards promoting the research on calibration-free BCIs.