Research Objectives List
Brain decoding algorithms constitute the ‘heart’ of a BCI system. Hence, the main objective here
is to conceive EEG decoding algorithms by combining recent advances in the fields of neuroscience and
Machine Learning. Beyond the conceptualization of novel decoding schemes, it is among our objectives to
achieve a deeper understanding of the neural processes that are related to the imagined speech and exploit
them towards developing reliable and effective Machine Learning algorithms.
The objective here is to establish the decoding schemes for an imagined speech BCI system that
will be able to incrementally learn how to decode new classes (e.g., new imagined words/phonemes/syllables)
using a small number of additional trials, without compromising the robustness of the existing vocabulary.
The main scope of this objective is to initially study the neural activations when one imagines
words with identical meaning but in different languages (e.g., ‘no’ and ‘ochi’; the greek word for no). Then, to
establish a framework that uncovers potential interconnections between such activation patterns and apply transfer learning approaches [Wan et al., 2021] that deal with the aforementioned interconnections.
The main concept of this objective is to create a publicly available EEG-based dataset, oriented
towards the imagined speech paradigm. The dataset will be formulated accordingly in order to facilitate a
benchmarking framework, hence, it will be accompanied by carefully crafted evaluation metrics.