Offered

FL is a new methodology providing privacy in learning by sharing only model parameters rather than the data directly with the server. In this project, we aim to apply well-known FL techniques to wearable data collected from university students and workers and compare them with traditional ML techniques. The dataset includes activity, sleep, and stress-related information.

Note that this project will be co-supervised by Suzan Ece Ada, Phd student in Colors Lab.

In robotic, simulators are a requirement for improvement of the algorithms without causing any harm, so existing robots should be transferred to asimilasyon environment. We need to use KIT Gripper with ROS + Gazebo environment with position controller. Also, existing robot should have similar inputs with the simulations. So, we need a ROS Wrapper which takes data from real robot and converts it to required ROS messages. After environment development, the system will be tested via Human - Robot cooperative game.

Training machine learning algorithms on resource-constrained mobile and wearable devices, particularly DL algorithms, is challenging and sometimes even impossible due to the limited computation power, storage, and, most importantly, the battery. TensorflowLite is a popular platform for optimizing deep learning architectures to be deployed on mobile devices. In this project, the focus is on human activity recognition with the motion sensors embedded in smartphones.

Although it is possible to generate realistic images, GAN models often lack fine-grained control over the generated images. Recently, several methods have been proposed to provide fine-grained control over latent space. Most of these works allow us to find domain-independent and general directions such as rotation, zoom-in or color change, while others proposed techniques to explore domain-specific directions such as changing age, gender, or expression on facial images.