Bayesian dynamic stopping for c-VEP BCIs

Context

While many conventional BCI systems adhere to fixed trial lengths, allowing users to issue commands only at predetermined intervals and leveraging a constant amount of EEG data, more sophisticated methods have emerged to optimize trial duration dynamically. Dynamic stopping methods enhance BCI system speed by deciding at any moment whether to output a result or wait for more information. Such approach leverages trial variance, allowing good trials to be detected earlier, thereby speeding up the process without significantly compromising accuracy. Existing dynamic stopping algorithms typically optimize measures such as symbols per minute (SPM) and information transfer rate (ITR). However, these metrics may not accurately reflect system performance for specific applications or user types. Moreover, many methods depend on arbitrary thresholds or parameters that require extensive training data. We have developed a model-based approach that takes advantage of the analytical knowledge that we have about the underlying classification model. By using a risk minimization approach, our model allows precise control over the types of errors and the balance between precision and speed. The risk minimization is performed through optimizing Bayes criterion and therefore we call the method Bayesian Dynamic stopping or in short BDS. The method is developed and tested on offline data sets. There are however a few interesting directions to investigate.

Image credit: S Ahmadi, P Desain, J Thielen, A Bayesian dynamic stopping method for evoked response brain-computer interfacing - arXiv preprint arXiv:2406.11081, 2024

Research question

• Performance comparison of dynamic stopping methods in c-VEP paradigm through literature analysis of dynamic stopping methods proposed for c-VEP or adaptable to c-VEP. This project requires investigating the generalizability of various dynamic stopping methods developed for different paradigms and understanding to what extent these methods have the capacity to be adapted to different paradigms.

• Comparing the performance and user experience of Bayesian dynamic stopping method with a few baseline methods in an online experiment.

• Investigating the effect of the size of training data on the performance of Bayesian dynamic stopping method.

• The current implementation of the Bayesian dynamic stopping is based on inner product as the similarity score between the predicted and observed responses. However, correlation is a more comon similarity score often used both in dynamic stopping and classification. Comparison between the performcance obtained with inner produc versus correlation as similaroty score on a few baseline stopping methods indicate slight preference for correlation. The question to investigate is how can the Baysian dynamic stopping be adapted to use correlation as similarity score and whether such adaptation can improve the performace.

Skills / background required

• Very proficient in Python
• Knowledge/interst in Brain-computer interfacing