HCC: Medium: Optimizing Interactive Machine Learning Tools to Support Plant Scientists using Human Centered Design

了解植物的结构和功能，特别是根随时间的变化，对于了解植物如何适应不断变化的气候和确保可持续生产至关重要。这方面的前沿进展得益于传感器可用数据的多样性和绝对数量的大幅增加。例如，植物科学使用被称为迷你加速器(MR)系统的传感器技术来研究根的发育。这些传感器通过放置在土壤中的摄像头以透明管的形式捕捉植物根系的彩色图像。由于需要人类对数据进行解释，准备这些图像用于科学研究需要大量的时间、人力和精力。机器学习算法可以自动化一些准备工作，但我们不知道如何设计一个系统来帮助人类和机器学习算法最好地协同工作。该项目开发方法和工具，以支持植物科学家(不同背景和专业水平，包括青年和其他新手)与机器学习一起工作，以更好地利用核磁共振系统。成果将包括人类图像标记者的交互式机器学习体验方面的进展，以及参与标记和作为科学家的自我认同之间的关系。该项目将对植物科学和其他领域的基于传感器的数据科学产生广泛的影响，解决多个具有全球重要性的问题。美国仍然缺乏正在培训的科学家，该项目将推进和评估吸引更多学生进入科学领域的努力。该项目中的科学教育项目让青年参与设计人机系统，可以帮助来自边缘背景的青年学习科学是如何工作的，并帮助他们将自己视为未来的科学家。该项目提供了所需的工具，以显著减少由磁共振系统生成的植物根系数据的分析瓶颈，并从长远来看，使可能具有重大全球重要性的更大规模的基于磁共振的研究成为可能。该项目的重点是开发交互式机器学习工具，旨在使用以人为中心的设计方法支持不同专业水平的植物科学家。为了实现这些目标，该项目对混合方法的结果进行三角测量，包括对经验丰富的标签师进行实验室研究，与来自不同背景的利益相关者共同参与设计研讨会，以及与佛罗里达州4-H合作伙伴计划的夏季参与式科学体验。这些实验室研究有助于对人类标注者行为(如标注数量和质量)如何影响机器学习算法性能以及反之亦然的新理解。参与式联合设计研讨会的重点是设计交互式机器学习数据可视化和标记工具，基于对具有不同专业知识的植物科学家的以人为中心的理解，包括作为最终用户的科学家、新兴科学家和非科学家，包括青年和成年人。最后，暑期科学经验告诉我们如何将这种方法扩展到更广泛的领域和用户群体，而不是那些年轻时传统上从事STEM的人。该项目将促进植物科学中磁共振系统数据分析的更高吞吐量，使未来能够对农业和自然生态系统的生产力、可持续性和弹性产生影响。它还将影响科学中以人为中心的机器学习的总体吞吐量。该项目的方法也将推广到其他类似的标记领域，如人体解剖学(血管、神经元)或水文学(河流三角洲、海岸线)。通过与4-H的伙伴关系让边缘化的年轻人参与进来，也会增加国家未来的STEM劳动力。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Understanding the structure and function of plants, especially roots as they change over time, is essential to understand how plants adapt to changing climates and ensure sustainable production. Cutting-edge advances in this understanding are benefitting from huge increases in the diversity and sheer amount of data available from sensors. For example, plant science uses sensor technology called minirhizotron (MR) systems to study root development. These sensors capture color images of plant roots through cameras placed into the soil in clear tubes. Preparing these images to be used in scientific research requires enormous amounts of time, labor, and effort, due to the need for human interpretation of the data. Machine learning algorithms can automate some of the preparation, but we do not know how to design a system to help humans and machine learning algorithms work best together. This project develops methods and tools to support plant scientists (of varying backgrounds and expertise levels, including youth and other novices) working in tandem with machine learning to better utilize MR systems. Outcomes will include advances in both the interactive machine learning experience for human image labelers, as well as the relationship between participating in labeling and self-identification as scientists. The project will have broad implications for sensor-based data science in plant science and beyond, addressing multiple issues of global importance. The U.S. continues to experience a shortage of scientists-in-training, and the project will advance and evaluate efforts to draw more students into science. Science education programs in this project, which involve youth in designing the human-machine system, can help youth from marginalized backgrounds learn how science works and help them see themselves as future scientists. This project provides the tools needed to significantly reduce the analysis bottleneck of the plant root data generated by MR systems and, in the long term, enable larger-scale MR-based studies that may have significant global importance. The focus of this project is to develop interactive machine learning tools targeted to support plant scientists of varying expertise levels using a human-centered design approach. To accomplish these goals, this project triangulates findings from mixed methods, including laboratory studies of experienced labelers, participatory co-design workshops with stakeholders from diverse backgrounds, and summer participatory science experiences with Florida 4-H partner programs. The laboratory studies contribute new understanding of how human labeler behavior (such as annotation quantity and quality) affects machine learning algorithm performance, and vice versa. The participatory co-design workshops focus on designing interactive machine learning data visualization and labeling tools based in a human-centered understanding of plant scientists of varying expertise, including scientists, emerging scientists, and non-scientists, both youth and adults, as end users. Finally, the summer science experiences inform on how to scale this approach to broader domains and user populations beyond those traditionally engaged in STEM as youth. This project will facilitate higher throughput in the analysis of MR systems data in plant science, enabling future impacts to productivity, sustainability, and resilience of agricultural and natural ecosystems. It will also impact the throughput of human-centered machine learning in science in general. Methods from this project will also generalize to other similar labeling domains, such as human anatomy (blood vessels, neurons) or hydrology (river deltas, coastlines). Involving marginalized youth through partnerships with 4-H also grows the nation’s prospective STEM workforce.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.