Masterthesen für Bioinformatik
Saccade BundlesThis project is about applying (and probably adjusting) a linear time clustering algorithm for brain fiber activity to work with saccadic trajectories. The result will be a clustering of saccades (= saccade bundle). The algorithm will be applied to eye-tracking data recorded while viewing fine art and while driving. These bundles could give insight into the perception of art (e.g. do saccades follow composition lines) and can be used to quantify saccadic distributions and traversals towards different interesting areas for driving (e.g. control gaze towards the rear mirror). The algorithm will be included in our eye-tracking data analysis software EyeTrace. Ansprechpartner: Dr. Thomas Kübler Automatisierter Scanpath VergleichBeschreibung:
Beide Methoden zeichne sich durch massive Einschränkungen und sehr begrenzte Anwendbarkeit aus: Sie funktionieren generell nur bei einfachen Scanpaths und unbewegtem Betrachter gut. Deshalb ist die Analyse von Hand noch immer sehr verbreitet, obwohl hierdurch Objektivität verloren geht und viel Zeit nötig ist. Aufgabenstellung: Ansprechpartner: Jun.-Prof. Dr. Enkelejda Kasneci, Dr. Thomas Kübler Klassifikation emotionaler Zustände durch HerzratenvariabilitätBeschreibung: Der Ausdruck und das Verstehen von Emotionen stellen nicht nur Überleben sondern auch unser Wohlbefinden im täglichen Leben sicher. Menschliche Kommunikation und damit Entscheidungsfindungen basieren stark auf Emotionen. Wissenschaftler erforschen diesen Sachverhalt im Feld Mensch-Computer Interaktion,umdieInteraktionzwischenMenschen und elektronischen Gera ?ten zu verbessern. Mit dem Aufkommen von kostengünstiger mobiler Hardware, welche es erlaubt die Herzaktivität oder andere körperliche Signale aufzuzeichnen, ist das Interesse solche Informationen in Alltagsgerätschaften wie Smartphones zu integrieren stetig gestiegen. Neben der quantified-self Bewegung hat der wissenschaftliche Aspekt automatisiert körperliche Signale in Kommunikationshilfen für bewegungseingeschränkte Menschen zu integrieren hohe Wichtigkeit. Ziele: Die Ziele dieser Thesis sind zweifältig. Analyse bereits aufgenommener Herzaktivitätdaten durch Herzratenvariabilität bezogen auf drei unterschiedliche emotionale Zustände, i.e. angenehm, neutral, unangenehm. Automatisierte Klassifikation dieser Merkmale in geeignetem maschinellem Lernansatz. Ansprechpartner: Dirk Tassilo Hettich Erweiterung des c-VEP Brain-Computer Interface (BCI) durch einen non-control stateBeschreibung: Ein Brain-Computer Interface (BCI) ist eine Schnittstelle zwischen dem menschlichen Gehirn und einem Computer und soll es körperlich gelähmten Patienten möglich machen einen Computer über reine Gehirnaktivität zu steuern. Das in Tübingen entwickelte c-VEP BCI ist das momentan schnellste nicht-invasive BCI System.
Auch wenn das c-VEP BCI momentan schon sehr schnelle und genaue Bedienung ermöglicht, so läuft es momentan noch in einem synchronen Modus, in dem nach einem festgelegten zeitlichen Muster eine Eingabe erwartet wird und die Person die das BCI steuert somit keine Pausen machen kann. Ziel dieser Arbeit soll es sein die Bedienung des c-VEP BCIs auch in einem asychronen Modus zu ermöglichen. Das BCI soll somit non-control states erkennen, also ob der Benutzer gerade das BCI bedienen will oder ob er gerade eine Pause macht. Im Rahmen der Arbeit sollen Methoden zur Erkennung des non-control states die in anderen BCIs Anwendung finden auf das c-VEP BCI übertragen werden und deren Genauigkeit an vorhandenen c-VEP Daten simuliert werden. Abhänging von den erzielten Ergebnissen kann eine Erkennung des non-control states auch in die vorhandene BCI Software eingebaut werden und an Probanden getestet werden.
Anforderungen: Interesse an Neurowissenschaften, gute Programmierkentnisse und Erfahrung in Matlab sind von Vorteil. Ansprechpartner: Dr. Martin Spüler Using spiking neural networks to simulate the effect of BCI-TrainingDescription: A Brain-Computer Interface (BCI) is a tool that allows to control a computer by brain activity only. Its main application is to establish communication for paralyzed patients, but BCIs are also used for stroke rehabilitation and to improve recovery in those patients. Especially in the latter case, the training effects during BCI usage play an important role since they induce neural plasticity and thereby help recovery in those patients. While it is a fact that using a BCI induces training effects and thereby neural plasticity, it is unknown how the BCI affects neural plasticity and how the BCI can be designed to optimize training effects.
Task: The idea behind this thesis is to simulate the effects of BCI training by using a spiking neural network. Chadderon et al. (2012, Plos One) have shown that spiking neural networks with reinforcement learning can be used to learn motor control of a simulated limb. A similar model should be implemented in the course of the thesis. Based on this spiking neural network model, an EEG recording should be simulated and used to control a BCI application. The output of the BCI can be used to give sensorimotor feedback to the neural network and thereby enable the neural network to learn better BCI control and thereby simulate the effects of BCI training. This model can be further used to evaluate how different BCI algorithms influence the learning and how they have to be designed to optimize learning. Ansprechpartner: Dr. Martin Spüler Using Spiking Neural Networks to model the effects of stroke on the human motor systemDescription: Stroke is one of the leading causes for disability. In the last years, Brain-Computer Interfaces (BCIs) have emerged as a tool to translate brain activity and it has been shown that BCIs can successfully used for stroke rehabilitation and to improve recovery in those patients. The training effects during BCI usage play an important role since they induce neural plasticity and thereby help recovery in those patients. While it is a fact that using a BCI induces training effects and thereby neural plasticity, it is unknown how the BCI affects neural plasticity and how the BCI can be designed to optimize training effects. To be able to simulate the effects of BCI training in stroke patients, a computational model is needed. Task: Chadderon et al. (2012, Plos One) have shown that spiking neural networks with reinforcement learning can be used to learn motor control of a simulated limb. An improved version of this network has been implemented in our lab. The focus of this work lies on extending the simulation of the simulated limb to be more realistic (including different muscles), adjust the neural network to be able to learn motor control of the more complex limb and investigate how effects of stroke can be modeled (by removing certain connections or neural populations). Requirements: High motivation, Good programming skills, experience with Matlab and a neuroscientific background is beneficial. Ansprechpartner: Dr. Martin Spüler Anwendung bekannter Algorithmen zur Klassifikation von Augenbewegungen aus Messdaten von Realfahrten bei der Daimler AGBeschreibung:
Aufgabenstellung: Ansprechpartner: Jun.-Prof. Dr. Enkelejda Kasneci |