CAREER: Autonomous Wearable Computing for Personalized Healthcare

职业：用于个性化医疗保健的自主可穿戴计算

• 批准号: 1750679
• 负责人: Hassan Zadeh
• 金额: $ 51.57万
• 依托单位: Washington State University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-05-15 至 2022-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1750679&HistoricalAwards=false
• 关键词: CAREER; Autonomous; Wearable; Computing; Personalized

Wearables are poised to transform health and wellness through automation of cost-effective, objective, continuous, real-time, and remote health monitoring and interventions. These technologies utilize machine learning algorithms to detect important health events and to predict impending medical complications. Currently, however, machine learning algorithms for these systems are designed based on sensor data that are collected and labeled/annotated in controlled environments such as laboratory settings and clinics. This process of data collection, data annotation, and algorithm training has created great impediments to scalability of wearable systems because (1) collecting and labeling sufficiently large amounts of sensor data is a time consuming, labor-intensive, expensive, and often infeasible task; and (2) wearables are deployed in highly dynamic environments of the end-users whose physical, behavioral, social, and environmental context undergo consistent changes. Such changes result in drastic decline in the accuracy of the machine learning algorithms trained in controlled environments. Therefore, it is important to develop autonomously reconfigurable machine learning algorithms as wearable sensors, settings in which they are utilized, and their configuration changes. This project introduces computational autonomy as an overarching solution for training accurate machine learning algorithms, without human supervision, in highly dynamic, unpredictable, and uncontrolled settings. The successful conclusion of this work will enable future wearables to learn in-situ autonomously, operate in-the-wild reliably, and adapt to the changing context of their users automatically. This project will develop foundations of computational autonomy for wearable-based health monitoring and interventions through the following research objectives: (1) investigating methods of automatic and autonomous labeling of sensor data in a new setting based on labeled sensor data collected in a different setting by designing combinational optimization methodologies for cross-subject, cross-context, cross-platform, and cross-modality sensor data mapping; (2) developing non-parametric label refinement algorithms to reliably infer labels in a new setting based on uncertain and sporadic knowledge obtained from another, potentially unreliable and heterogeneous, sensor; (3) exploring methodologies for training machine learning algorithms that are robust to unknown parameters of a source sensor and adaptive to dynamically changing signal attributes of the new setting; and (4) validating the developed algorithms and tools through both in-lab experiments and in-the-wild user studies.This interdisciplinary project will not only address the technical challenges in developing highly performant wearable systems but will also enable actual monitoring of a variety of populations. The work has major broader impacts on conducting high-precision chronic disease management and on the availability of wearable-based consumer applications. This has the potential to lead to the development of products around the concept of computational autonomy and its use in automation of health management, as well as, applications yet to be envisioned. The interdisciplinary nature of this work will provide unique opportunities for integrated research and education. To this end, the educational objectives will focus on developing a new ambassador program to increase interest of underrepresented minority community college students in Science, Technology, Engineering and Math (STEM) careers in general and in computer science and engineering careers in particular, developing a novel patron program to improve retention of transferred underrepresented minority students through student and parental exposure to wearable-based health monitoring research, engaging undergraduate students in research, and establishment of an interdisciplinary research-based curriculum on computational autonomy. All the data produced over the course of this project, including design methodologies, software algorithms and tools, experimental data, publications, and curriculum will be made publicly available at http://epsl.eecs.wsu.edu/. The data will be stored and hosted on local servers and replicated on external public web servers.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.

可穿戴设备有望通过自动化的成本效益，客观，连续，实时和远程健康监测和干预来改变健康和健康。这些技术利用机器学习算法来检测重要的健康事件并预测即将发生的医疗并发症。然而，目前，这些系统的机器学习算法是基于在受控环境（如实验室设置和诊所）中收集和标记/注释的传感器数据设计的。这种数据收集、数据注释和算法训练的过程对可穿戴系统的可扩展性造成了很大的障碍，因为（1）收集和标记足够大量的传感器数据是一项耗时、劳动密集、昂贵且通常不可行的任务;以及（2）可穿戴设备被部署在终端用户的高度动态环境中，终端用户的物理、行为、社交和环境背景不断变化。这种变化导致在受控环境中训练的机器学习算法的准确性急剧下降。因此，重要的是开发自主可重构机器学习算法作为可穿戴传感器，它们被利用的设置以及它们的配置变化。该项目引入了计算自主性，作为在高度动态、不可预测和不受控制的环境中训练准确机器学习算法的总体解决方案，无需人工监督。这项工作的成功完成将使未来的可穿戴设备能够在现场自主学习，在野外可靠地运行，并自动适应用户不断变化的环境。该项目将通过以下研究目标为基于可穿戴设备的健康监测和干预建立计算自主性基础：（1）通过设计跨学科，跨上下文，跨平台和跨模态传感器数据映射的组合优化方法，基于在不同环境中收集的标记传感器数据，研究在新环境中自动和自主标记传感器数据的方法;（2）开发非参数标签细化算法，以基于从另一个潜在不可靠和异构的传感器获得的不确定和零星的知识来可靠地推断新设置中的标签;（3）探索用于训练机器学习算法的方法，所述机器学习算法对源传感器的未知参数是鲁棒的并且适应于新设置的动态变化的信号属性;（4）通过实验室实验和野外用户研究验证所开发的算法和工具。这个跨学科项目不仅将解决开发高性能可穿戴系统的技术挑战，还将实现对各种人群的实际监测。这项工作对进行高精度的慢性病管理和基于可穿戴设备的消费者应用的可用性产生了更广泛的影响。这有可能导致围绕计算自主性概念的产品开发及其在健康管理自动化中的应用，以及尚未设想的应用。这项工作的跨学科性质将为综合研究和教育提供独特的机会。为此，教育目标将侧重于开发一个新的大使计划，以提高代表性不足的少数民族社区大学生对科学，技术，工程和数学（STEM）职业的兴趣，特别是计算机科学和工程职业，开发一种新的赞助计划，通过学生和家长接触可穿戴设备，基于健康监测的研究，从事本科生的研究，并建立一个跨学科的研究为基础的计算自主性课程。在这个项目的过程中产生的所有数据，包括设计方法，软件算法和工具，实验数据，出版物和课程将在http://epsl.eecs.wsu.edu/上公开。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

LabelMerger: Learning Activities in Uncontrolled Environments

LabelMerger：不受控制的环境中的学习活动

• DOI: 10.1109/transai46475.2019.00019
• 发表时间: 2019
• 期刊: 2019 First International Conference on ​Transdisciplinary AI (TransAI
• 作者: Mirzadeh, Seyed Iman;Ardo, Jessica;Fallahzadeh, Ramin;Minor, Bryan;Evangelista, Lorraine;Cook, Diane;Ghasemzadeh, Hassan
• 通讯作者: Ghasemzadeh, Hassan

CyclePro: A Robust Framework for Domain-Agnostic Gait Cycle Detection

• DOI: 10.1109/jsen.2019.2893225
• 发表时间: 2019-05-15
• 期刊: IEEE SENSORS JOURNAL
• 影响因子: 4.3
• 作者: Ma, Yuchao;Ashari, Zhila Esna;Ghasemzadeh, Hassan
• 通讯作者: Ghasemzadeh, Hassan

Dropout as an implicit gating mechanism for continual learning

Dropout 作为持续学习的隐式门控机制

• DOI: 10.1109/cvprw50498.2020.00124
• 发表时间: 2020
• 期刊: In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR
• 作者: Mirzadeh, Seyed Iman;Farajtabar, Mehrdad;Ghasemzadeh, Hassan
• 通讯作者: Ghasemzadeh, Hassan

LabelForest: Non-Parametric Semi-Supervised Learning for Activity Recognition

LabelForest：用于活动识别的非参数半监督学习

• DOI: 10.1609/aaai.v33i01.33014520
• 发表时间: 2019
• 期刊: Proceedings of the AAAI Conference on Artificial Intelligence
• 作者: Ma, Yuchao;Ghasemzadeh, Hassan
• 通讯作者: Ghasemzadeh, Hassan

Linear Mode Connectivity in Multitask and Continual Learning

• 发表时间: 2020-10
• 期刊: ArXiv
• 作者: Seyed Iman Mirzadeh;Mehrdad Farajtabar;Dilan Gorur;Razvan Pascanu;H. Ghasemzadeh