If you are interested in any of these topics, please send me a brief email with your previous experiences related to the topic and your (programming) skills by March 9th, 2020.Topic: Autoencoders for feature extraction in wireless cognitive load inference
Knowing a user's cognitive load can greatly enhance the number of ubiquitous computing applications. For instance, a mobile phone aware of its user's high cognitive engagement could postpone a notification for later, a mobile game could sense when a level is becoming to easy for the player, and a smart car could detect when a driver is falling asleep. In a previous MS thesis we have developed a wireless radar for cognitive load inference. Breathing and heartbeats change with cognitive load and such a radar can detect minute movements corresponding to the above vital signals. Yet, machine learning models that use features hand crafted from the wireless signals exhibit a relatively poor accuracy. In this thesis you will investigate the use of autoencoders to automatically extract informative features from the collected signals and push the limits of wireless cognitive load inference.
NOTE: Python proficiency is essential. Previous knowledge of deep learning is desired. Must be written in English.Topic: Adaptive mobile video resolution playback based on human activity recognition
Designing and implementing an Android app consisting of an open source video player (e.g. NewPipe) and an open source human activity classifier (e.g. Human-Activity-Recognition-Keras-Android). The video player should be configured to enable either a manual resolution (selectable by the user) or automatic resolution based on the current mobility state. In the automatic mode, the input from the phone’s accelerometer is applied to the classifier which determines the mobility state; the system adapts the resolution based on this state (e.g. when sitting the resolution is higher, for example 720p, when jogging the system sets a lower resolution, e.g. 360p). The goal of the approach is to lower the energy consumption, yet still satisfy the user's needs. After the implementation, the app is evaluated with at least 20 volunteers and its ability to satisfy users' requirements and preserve battery are assessed.
NOTE: Android app development; User study; Basic machine learning. Must be written in English.
At the moment I have no funded positions for postdocs and PhD students. However, highly-motivated postdoctoral candidates who want to apply for EU Marie Sklodowska-Curie actions (MSCA) fellowships are encouraged to contact me. MSCA fellowships are an excellent opportunity for researchers who have recently obtained their PhD to build their independent career. The fellowships are open to any nationals and provide a very generous funding for a research and her/his family to move to Slovenia. The University of Ljubljana provides strong support for crafting a successful MSCA application. Please read the following webpage for more information, and email me (Veljko.Pejovic [at] fri.uni-lj.si) with your CV and a breif ressearch outline if interested.