Collaborative Multi-Expert Active Learning for Mobile Health Monitoring: Architecture, Algorithms, and Evaluation.

Mobile health monitoring plays a central role in the future of cyber physical systems (CPS) for healthcare applications. Such monitoring systems need to process user data accurately. Unlike in other human-centered CPS, in healthcare CPS, the user functions in multiple roles all at the same time: as an operator, an actuator, the physical environment and, most importantly, the target that needs to be monitored in the process. Therefore, mobile health CPS devices face highly dynamic settings generally, and accuracy of the machine learning models the devices employ may drop dramatically every time a change in setting happens. Novel learning architecture that specifically address challenges associated with dynamic environments are therefore needed. Using active learning and transfer learning as organizing principles, we propose a collaborative multiple-expert architecture and accompanying algorithms for the design of machine learning models that autonomously adapt to a new configuration, context, or user need. Specifically, our architecture and its constituent algorithms are designed to manage heterogeneous knowledge sources or experts with varying levels of confidence and type while minimizing adaptation cost. Additionally, our framework incorporates a mechanism for collaboration among experts to enrich their knowledge, which in turn decreases both cost and uncertainty of data labeling in future steps. We evaluate the efficacy of the architecture using two publicly available human activity datasets. We attain activity recognition accuracy of over 85 % (for the first dataset) and 92 % (for the second dataset) by labeling only 15 % of unlabeled data.

Older People with Access to Hand-Held Devices: Who Are They?

INTRODUCTION: Multiple comorbid conditions among older patients require frequent physician office and emergency room visits, at times leading to hospitalization. In recent years, mobile health (m-health) systems utilizing hand-held devices (e.g., smartphones) have been developed, which could be used for health-related interventions. This study investigates sociodemographic and clinical characteristics of individuals who have or have not accessed Internet via hand-held devices. MATERIALS AND METHODS: Adults older than 65 years of age who participated in the Health Tracking survey of the Pew Internet and American Life Project in 2012 were included in the analysis. Data were analyzed for prevalence of Internet access via hand-held devices and differences in sociodemographic and clinical characteristics. Different online health information seeking behavior is also reported. RESULTS: In the weighted sample size of 3,116 responses, 472 (15.1%) had access to Internet via hand-held devices. Those with such an access were younger and had higher income and education and better overall quality of life and quality of life at the time of answering the survey. They were more likely to be female and married or living as married. Those with diabetes or significant change in physical condition in the prior year were less likely to have such an access. In the multivariate analysis, older or diabetic individuals had lower probability of such access. Higher likelihood of access was associated with higher income and education, being married, female gender, better quality of life, higher number of comorbid illnesses, and emergency room visit or hospital admission in the last 12 months. CONCLUSIONS: Investigators should pay attention to sociodemographic and clinical disparities of older adults to develop feasible m-health interventions.

Synthetic Sensor Data Generation for Health Applications: A Supervised Deep Learning Approach.

Recent advancements in mobile devices, data analysis, and wearable sensors render the capability of in-place health monitoring. Supervised machine learning algorithms, the core intelligence of these systems, learn from labeled training data. However, labeling vast amount of data is time-consuming and expensive. Moreover, sensor data often contains personal information that a user may not be comfortable sharing. Therefore, there is a strong need to develop methods for generating realistic labeled sensor data. In this paper, we propose a supervised generative adversarial network architecture that learns from feedback from both a discriminator and a classifier in order to create synthetic sensor data. We demonstrate the effectiveness of the architecture on a publicly available human activity dataset. We show that our generator learns to output diverse samples that are similar but not identical to the training data.

Personalized Human Activity Recognition using Wearables: A Manifold Learning-based Knowledge Transfer.

Human activity recognition (HAR) is an important component in health-care systems. For example, it can enable context-aware applications such as elderly care and patient monitoring. Relying on a set of training data, supervised machine learning algorithms form the core intelligence of most existing HAR systems. Meanwhile, the accuracy of an HAR model highly depends on the similarity between the training and the operating context. Therefore, there is a need for developing machine learning algorithms that can easily adapt to the operating context at hand. In this paper, we propose a cross-subject transfer learning algorithm that links source and target subjects by constructing manifolds from feature-level representation of the source subject(s). Our algorithm assigns labels to the unlabeled data in the current context using the manifold learned from the source subject(s). The newly labeled data is used to develop a personalized HAR model for the current context (i.e., target subject). We demonstrate the efficacy of the algorithm using a publicly available dataset on HAR. We show that the proposed framework improves the accuracy of activity recognition by up to 24%.

Toward seamless wearable sensing: Automatic on-body sensor localization for physical activity monitoring.

Mobile wearable sensors have demonstrated great potential in a broad range of applications in healthcare and wellness. These technologies are known for their potential to revolutionize the way next generation medical services are supplied and consumed by providing more effective interventions, improving health outcomes, and substantially reducing healthcare costs. Despite these potentials, utilization of these sensor devices is currently limited to lab settings and in highly controlled clinical trials. A major obstacle in widespread utilization of these systems is that the sensors need to be used in predefined locations on the body in order to provide accurate outcomes such as type of physical activity performed by the user. This has reduced users' willingness to utilize such technologies. In this paper, we propose a novel signal processing approach that leverages feature selection algorithms for accurate and automatic localization of wearable sensors. Our results based on real data collected using wearable motion sensors demonstrate that the proposed approach can perform sensor localization with 98.4% accuracy which is 30.7% more accurate than an approach without a feature selection mechanism. Furthermore, utilizing our node localization algorithm aids the activity recognition algorithm to achieve 98.8% accuracy (an increase from 33.6% for the system without node localization).