Analysing Brain Signals
Are we in the future able to control technical devices using only our thoughts and by monitoring brain activity. The master projects below are to be undertaken together with neuroscience experts at Department of Psychology at UiO. In general, we are interested in developing models that use brain signals (mainly EEG) to either predict mental states or provide feedback back to the person about brain states. In the end, our ambition is to inform the development of prosthetic devices based on scalp EEG. This can be expressed in three related areas of research interests.
Some examples of projects:
Analysing Brain Signals for Control
Are we in the future able to control technical devices using only our thoughts and by monitoring brain activity. This master project is to be undertaken together with neuroscience experts at Department of Psychology at UiO. In general, we are interested in developing models that use brain signals (mainly EEG) to either predict mental states or provide some control functionality. In the end, our ambition is to inform the development of using thoughts in human-machine interfaces.
For this thesis project we would like to explore how machine learning techniques could be applied to the analysis of EEG signals, in order to better understand such signals, and ultimately apply them to real-time BCI applications, such as simple switch control or robot control. There are already examples in the research literature of robot control systems using brain signals. Some of these have been acquired by intracranial sensors, and while this allows for better precision in control, such systems are not very flexible and realistic for healthy people. Other systems based on scalp EEG signals have been proposed, many of these rely on the P300 event related potential (ERP), which can be consistently detected 300ms after a “surprising” event. This has the advantage of being a relatively reliable method, but on the other hand, the robot control suffers from long latency and low precision given discretization of the actions. It would thus be interesting to investigate other brain signal frequencies or events in order to reduce latency or get more proportional control.
The research in the thesis could take several directions. One direction could focus on setting up a physical system with basic functionality which allows for initial exploration of the real-time aspect and identifying possibilities of current approaches. Another direction could take a more analytical approach and focus mainly on machine learning methods for extracting information from the EEG signals. Simpler tasks could be considered as stepping stones towards robot control. E.g., one could focus on one-dimensional control problems.
Tasks for the project include:
- Doing a literature study of related approaches to the subject
- Develop and implement machine learning techniques for extracting information from EEG signals
- Possibly investigating a physical setup with real-time data collection and control/feedback
- Assess the performance of these methods in relevant use cases
- Writing the thesis report
- Brain signals converted into sound.Make a system that will take EEG signals as input and playback auditive information based on the signal. This can be implemented as a “make the brain sing” device or as biofeedback. One way would be to use brain signals to rise or lower pitch or volume or placement of sound in space or change a rhythm pattern.
- Prediction of rhythm.The project focuses around how the brain represents and predicts perceived rhythms. A subproject could be to develop a decoding system that can be used for classification purposes in research related to temporal prediction. We would particularly be interested in the signals associated with perception of and control over acceleration, deceleration and steady tempo rhythmic signals.
- Rhythm related mental state monitoring? We know that there are “rhythm like” oscillations in the human brain. We would like develop analysis tools that enable us to link “brain rhythms” to different types of behavior. Particularly we would like to identify different subsystems that are expressing rhythms in conditions where people relax and do nothing. We would particularly want to explore if machine-learning techniques serve as better classifiers than standard statistical approaches like independent component analysis (ICA).