H31: Conceptual and computational models of Brain-Computer Interfaces operation

Paper Submission for H31

SMC2018:H31 submission site (external site)

Abstracts

While promising for many applications, e.g., assistive technologies, gaming or adaptive human-computer interaction, Brain-Computer Interfaces (BCI) are still not robust enough to be used in practice, outside laboratories, for such applications. There is thus a need to improve BCIs’ robustness, i.e., to design BCIs that can be trusted to properly recognize the mental commands and/or mental states of their users. So far, most of BCI research has tried to address this problem either (1) by trials and errors, e.g., by trying out various classifiers, (2) according to heuristics, e.g., by selecting signal processing tools robust to the most likely source of variabilities, or feedbacks that are the most likely to improve users’ motivation. While such approaches have resulted in multiple improvements to BCI research and technology, they are mostly experimental. Since they are not grounded in theories and models, they may not be optimal. Unfortunately, there is a critical lack of models and theories dedicated to the understanding of BCI performance variabilities. For instance, there is no validated model of EEG signals’ non-stationarity and noise during BCI operation, nor a validated model of BCI user training and performance. Yet, such models would enable the community to design optimal signal processing algorithms and training approaches. For instance could be used to design EEG signals processing tools that optimally tackle EEG non-stationarity by being invariant to it. They could also enable us to optimally train BCI users to acquire BCI control through the optimization of the learning process. In this special session, we are thus seeking papers presenting and/or validating conceptual or computational models that can explain or predict BCI performances over time, context and users. We notably look for models explaining noise and non-stationarities in EEG signals during BCI control, as well as models explaining BCI user learning and performance. In other words, we want to start building a theoretical foundation of BCI operation, both at the machine (EEG signal processing) and the user levels (BCI user training), but also at the level of their interaction (e.g., co-adaptive BCI training).

Session Chairs