Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Thesis_Banerjee_OhioLink.pdf (20.87 MB)

ETD Abstract Container

Abstract Header

FedMyPath: Federated Learning-based Surface Classification using Continuous Data Streams from Wheelchair Users

Banerjee, Rochishnu
http://orcid.org/0000-0002-0114-2452
http://rave.ohiolink.edu/etdc/view?acc_num=miami1732276879210644

Abstract Details

2024, Master of Science in Computer Science, Miami University, Computer Science and Software Engineering.
Barriers such as steep inclines, rough surfaces, uneven or broken pavements, and other environmental factors pose mobility challenges for wheelchair users. The impact of these obstacles varies from person to person based on their specific needs and mobility aids. To study surface-induced vibrations, we propose a machine learning-based classification system to identify surface types from vibrational patterns captured via smartphone accelerometers and gyroscopes, using the MyPath data collection app. However, our previous centralized solution, MyPath, faced scalability and privacy challenges and incurred high network bandwidth for data transfer. In this research, we introduce FedMyPath, a distributed, privacy-preserving solution based on federated learning. In FedMyPath, individual users train a model using their locally held private data, sharing only model parameters and updates with a central server. This approach ensures data privacy, reduces communication overhead, and provides a scalable alternative. FedMyPath performs surface classification using continuous data streams from wheelchair users. We extensively evaluate the system for feasibility, performance, and adaptability to continuously generated vibration data collected from 10 surface types by 23 users. Our experimental results demonstrate that FedMyPath achieves a peak accuracy of 69.74% in the presence of highly non-IID data.
Vaskar Raychoudhury (Advisor)
Xianglong Feng (Committee Member)
Honglu Jiang (Committee Member)
64 p.
Computer Science

Recommended Citations

Citations

  • Banerjee, R. (2024). FedMyPath: Federated Learning-based Surface Classification using Continuous Data Streams from Wheelchair Users [Master's thesis, Miami University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=miami1732276879210644

    APA Style (7th edition)

  • Banerjee, Rochishnu. FedMyPath: Federated Learning-based Surface Classification using Continuous Data Streams from Wheelchair Users. 2024. Miami University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=miami1732276879210644.

    MLA Style (8th edition)

  • Banerjee, Rochishnu. "FedMyPath: Federated Learning-based Surface Classification using Continuous Data Streams from Wheelchair Users." Master's thesis, Miami University, 2024. http://rave.ohiolink.edu/etdc/view?acc_num=miami1732276879210644

    Chicago Manual of Style (17th edition)

miami1732276879210644
54
© 2024, all rights reserved.
This open access ETD is published by Miami University and OhioLINK.
Release 3.2.12