Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Personalized Deep Learning for IoT-Enabled Health Monitoring

Jia, Zhenge (2022) Personalized Deep Learning for IoT-Enabled Health Monitoring. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

PDF (ETD submitted version)
Submitted Version

Download (18MB) | Preview


Biomedical sensors have been widely utilized to perform long-term health monitoring on biosignals by being embedded into Internet-of-Things (IoT) devices. IoT-enabled health monitoring is increasingly considered to be a promising alternative to conventional analytical instruments due to their durability, low cost, and simplicity. While IoT-enabled health monitoring provides the capability of detecting diseases that occur sporadically and acutely, current detection methods still count on a variety of heuristic criteria with carefully selected features. Considerable domain expertise is demanded in the process of detection methods design and the detection parameters adjustment for better detection performance. Recently, deep learning is gaining more attention in the healthcare industry. The most significant advantage of deep learning is that it could automatically execute feature engineering with only labeled data, which results in a great reduction in the expertise involvement and manual work in conventional detection methods.

However, directly applying deep learning is not always feasible for IoT-enabled health monitoring. First, biosignals are highly variable among patients in terms of morphological characteristics due to individual differences. The detection performances of the pre-trained deep learning model would degrade significantly on some patients. Therefore, effective model personalization methods are in urgent need in patient-specific detection. Second, the deep model personalization process still requires an extensive amount of labeled data. In practice, for some applications, it is impractical to obtain adequate labeled samples due to the overwhelmed workload in manual labeling. Third, the data access is limited due to privacy concerns in certain health monitoring applications, where aggregating personal health data in a centralized server is strictly restricted.

To address the aforementioned challenges, this dissertation proposes several techniques to enable personalized deep learning for IoT-enabled health monitoring. First, a novel meta-learning algorithm and a prior knowledge incorporated learning approach are proposed to obtain a well-generalized model initialization and to regularize the personalization process with medical knowledge. Second, a system-level design is proposed to conduct self-supervised and on-device model personalization. Finally, we propose a personalized meta-federated learning method for distributed IoT health monitors to generate a patient-specific model through collaborative training without accessing personal health data.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Jia, Zhengezhenge.jia@pitt.eduzhj190000-0002-0554-3608
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairHu, JingtongJTHU@pitt.edu0000-0003-4029-4034
Committee MemberAkcakaya,
Committee MemberZhan,
Committee MemberMiskov-Zivanov,
Committee MemberSaba,
Date: 6 September 2022
Date Type: Publication
Defense Date: 24 June 2022
Approval Date: 6 September 2022
Submission Date: 6 June 2022
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 157
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Electrical and Computer Engineering
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Personalization, Deep learning, Health monitoring, IoT
Date Deposited: 06 Sep 2022 16:20
Last Modified: 06 Sep 2022 16:20


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item