Titulo Estágio
Reducing training time and increasing accuracy of Brain-Computer Interfaces through Transfer Learning
Local do Estágio
CISUC-DEI
Enquadramento
A Brain-Computer Interface (BCI) is a system that provides a direct communication between the brain and a computer or external device. In short, it must interpret brain activity and translate it into commands that can be used to control devices or programs.
Different types of neuroimaging modalities can be used to implement BCIs. The most common modality is electroencephalography (EEG), since it provides a portable, inexpensive, non-invasive solution to measure brain activity with high temporal resolution. Within this neuroimaging modality, there are several approaches to generate brain signals that can be interpreted and transformed into commands by the BCIs, namely event-related potentials (the most prominent being the P300), steady-state visual evoked potentials (SSVEP) or event-related synchronization/desynchronization (ERS/D) through mental imagery. The P300 approach uses an oddball paradigm where an infrequent stimulus of interest is presented in a sequence of frequent stimuli of non-interest. With this paradigm, a positive deflection of the EEG measured in the central and posterior parts of the scalp is observed approximately around 300 ms after the infrequent stimulus of interest is presented.
There are still many challenges facing P300-based BCIs to be used more broadly. Achieving portable and practical BCIs that are easy to setup and fast to calibrate is currently a research line of big interest, since it would favorably help the adoption of this new technology in everyday settings. However, different issues causing low robustness and reliability should be addressed for these systems to be used in real life. Indeed, due to high intra- and inter-subject variabilities, most BCIs require time-consuming calibrations to maximize their performance, which makes the creation of one-model-fits-all solutions difficult.
One possibility to overcome longer calibration times is to perform transfer learning, using deep learning classifiers pre-trained in larger datasets that are then just tuned for the participant, thus requiring less calibration data.
Objetivo
1) Explore and evaluate transfer learning techniques to reduce calibration times in P300-based BCIs
2) Improve classification accuracies of other event-related potential (ERP) paradigms, by transferring learning from P300 dataset
Plano de Trabalhos - Semestre 1
1. State-of-the-art review: ERPs, P300 classification, conventional and deep learning approaches, transfer learning
2. Identification and familiarization with datasets to use
3. Definition of the methodological approach
4. Writing of the Intermediate Report
Plano de Trabalhos - Semestre 2
5. Research and development of transfer learning approaches for P300 classification
6. Experimental analysis and reporting of the developed DL models
7. Research and development of transfer learning approaches for other ERP signals
8. Experimental analysis and reporting of the developed DL models
9. Writing of a scientific paper and dissertation
Condições
Access to a GPU-enabled mainframe for running the models.
Possibility of partial funding depending on the curriculum of the candidate.
Observações
Co-suppervision by Davide Borra, University of Bologna
Orientador
Marco António Machado Simões
msimoes@dei.uc.pt 📩