Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Learning Battery-Efficient Sampling Policies for Human Activity Recognition
You will use reinforcement learning to train an adaptive human activity recognition system for smartphone sensor data that is battery-efficient in continuous (24/7) monitoring of mobility behaviors.
Keywords: Machine Learning, Reinforcement Learning, Mobile Health, Human Activity Recognition
Continuous human activity recognition (HAR) with smartphone sensors (accelerometer, gyroscope, etc.) could greatly advance our understanding of the interdependencies of physical activities, mobility and health. However, a major issue in using smartphone sensors for continuous monitoring of physical activities is the large battery usage incurred by permanently keeping smartphone sensors engaged. We know, from a previous project, that large battery-efficiency gains (up to 80%) are possible with only a small decrease in accuracy, if we are able to develop smart sampling strategies. In this project, we will use reinforcement learning to automatically learn a battery-efficient sampling strategy.
Your work will be a key component in future smartphone-based big health data studies on the interdependencies of physical activities and health and, thus, help advance our understanding of some of the most complex diseases.
Skills in software development and machine learning are required for this project.
Continuous human activity recognition (HAR) with smartphone sensors (accelerometer, gyroscope, etc.) could greatly advance our understanding of the interdependencies of physical activities, mobility and health. However, a major issue in using smartphone sensors for continuous monitoring of physical activities is the large battery usage incurred by permanently keeping smartphone sensors engaged. We know, from a previous project, that large battery-efficiency gains (up to 80%) are possible with only a small decrease in accuracy, if we are able to develop smart sampling strategies. In this project, we will use reinforcement learning to automatically learn a battery-efficient sampling strategy.
Your work will be a key component in future smartphone-based big health data studies on the interdependencies of physical activities and health and, thus, help advance our understanding of some of the most complex diseases.
Skills in software development and machine learning are required for this project.
Not specified
Patrick Schwab (patrick.schwab@hest.ethz.ch),
Mobile Health Systems Lab (www.mhsl.hest.ethz.ch),
Institute of Robotics and Intelligent Systems,
Department of Health Sciences and Technology,
ETH Zurich
This project will be co-supervised by Prof. Walter Karlen from the Mobile Health Systems Lab, D-HEST and Prof. Otmar Hilliges from the Advanced Interactive Technologies Lab, D-INF (https://ait.ethz.ch/).
Patrick Schwab (patrick.schwab@hest.ethz.ch), Mobile Health Systems Lab (www.mhsl.hest.ethz.ch), Institute of Robotics and Intelligent Systems, Department of Health Sciences and Technology, ETH Zurich
This project will be co-supervised by Prof. Walter Karlen from the Mobile Health Systems Lab, D-HEST and Prof. Otmar Hilliges from the Advanced Interactive Technologies Lab, D-INF (https://ait.ethz.ch/).