CI-P: Toward Brain-Computer Interfaces that Adapt to User Cognitive State

脑机接口(BCI)对肢体残疾用户非常有帮助，允许他们通过控制大脑活动来提供输入，大脑活动是通过专门的硬件感知的。这类硬件正迅速变得不那么昂贵和引人注目，为开发面向更广泛受众的BCI铺平了道路。这个项目是关于开发一个基于BCI的基础设施来感知一个人的认知工作量，然后使用这些信息来构建计算机如何与人互动的结构。该基础设施将使围绕几种自适应界面的研究成为可能，包括使用大脑活动使课程变得更难或更容易的教育工具，以及评估数据可视化的可理解性和难度的方法。这项研究有可能通过改善一般人和残疾人的教育、数据分析和界面来影响社会。此外，首席研究员将利用基础设施和使能的研究来吸引传统上在计算机科学中代表性较低的群体的学生，并支持情感计算和数据可视化课程。资金支持必要的基础设施，以建立和测试实时智能地响应用户认知状态的自适应用户界面。特别是，PI将获得多通道频域fNIRS(功能近红外光谱)设备，并开发处理fNIRS信号以提取工作量信息所需的算法。对于fNIRS设备提供的16个通道中的每一个，将对信号进行预处理，以考虑运动伪影并得出有意义的特征(值得注意的是，基于文献的平均和线性回归斜率)。该团队计划使用个性化的基于支持向量机的模型来区分高和低工作负载状态，这是基于PI之前的工作，表明了这种方法的有效性。

Brain-computer interfaces (BCIs) are incredibly helpful for physically disabled users, allowing them to provide input through controlling their brain activity, which is sensed through specialized hardware. Such hardware is rapidly becoming less expensive and obtrusive, paving the way for developing BCIs for a wider audience. This project is about developing an infrastructure for BCI-based sensing of a person's cognitive workload, then using this information to structure how the computer interacts with the person. The infrastructure will enable research around several kinds of adaptive interface, including educational tools that use brain activity to make lessons harder or easier and ways to evaluate the comprehensibility and difficulty of data visualizations. The research has the potential to impact society by improving education, data analysis, and interfaces for both people in general and people with disabilities. Further, the lead researcher will use the infrastructure and enabled research to attract students from groups traditionally underrepresented in computer science, as well as to support courses on affective computing and data visualization.The funding supports the infrastructure necessary to build and test adaptive user interfaces that respond intelligently to user cognitive state in real-time. In particular, the PIs will acquire a multichannel frequency domain fNIRS (functional near infrared spectroscopy) device and develop the algorithms required to process fNIRS signals to extract workload information. Signals will be pre-processed to account for motion artifacts and to derive meaningful features (notably, mean and linear regression slope based no the literature) for each of the 16 channels provided by the fNIRS device. The team plans to use personalized support vector machine-based models to distinguish between high and low workload states, based on prior work by the PIs that shows the effectiveness of this approach.