Generates a linear-nonlinear Bernoulli model for predicting discrete events, useful for EEG decoding
This code can be used to generate a regularized event-related potential (ERP) model, which can be used to create a time-varying event probability based on EEG data. The model assumes that the likelihood of events is determined by dot produced between the ERP and the EEG followed by a nonlinearity defined by a sigmoid function to make the probability Bernoulli-distributed. The fit of the time-varying probability is quantified using the log-likelihood, which can be compared to a null distribution based on shuffled and permuted data to evaluate the goodness of fit.
create_erp_regularize
-- uses 10-fold cross-validation to create an regularized ERP model with ridge regularizationpredict_eeg_events_idx
-- computes a time-varying probability of events in a trial using a given ERP model and nonlinearity
This model was used to quantify the predictability of phonemes, vowels, and consonants from EEG recorded during continuous listening to speech. A conference paper based on this work can be found here, please cite it if you intend to use these functions:
Zuk NJ, Di Liberto GL, Lalor EC (2019). Linear-nonlinear Bernoulli modeling for quantifying temporal coding of phonemes in continuous speech. Conference on Cognitive Computational Neuroscience, 13-16 September, Berlin, Germany. doi: 10.32470/CCN.2019.1192-0