Electroencephalography (EEG) signals represent the summed electrical activity of neural processes in the brain. Isolating specific neural responses within EEG recordings is challenging due to the inherently low signal-to-noise ratio. For this reason, a common approach is to average multiple stimulus-locked trials in order to extract the corresponding event-related potential (ERP). However, trial averaging suppresses intra-subject variability and removes dynamic information that may carry physiologically meaningful signatures of cognitive state. In contrast, single-trial analysis enables real-time interpretation, preserves temporal variability, and supports personalized modeling. This approach, however, requires methods capable of detecting weak, nonlinear, and noise-contaminated neural responses without relying on repeated averaging.