Abstract:

Neural activity is inherently noisy. To attenuate this noise, data is often time locked to an event and averaged over many repetitions. Assuming that the noise is uncorrelated to the event onsets, trials can be linearly averaged and given enough repetitions, this approach will dampen the noise and recover the underlying signal.

We and others recently showed over several domains that this assumption is not warranted:

Often, 

1) ERPs overlap in time. For instance when stimulus and button presses are close together as in decision making (Frömer et al. 2023 [1]), or when subsequent eye-movements overlap (Gert et al. 2022 [2])

2) predictors relate in non-linear ways. For instance eye-movement parameters in Dimigen et al. 2021 [3].

In this talk, I will present the regression ERP framework, based on multiple regression, linear deconvolution and Generalized Additive Modelling (GAMs) - which allows us to adress complex paradigms ranging from eye-movements, mobile EEG, reading, decision making - and, surprisingly, sheds new light on established paradigms like the P300 and CPP. I will base my talk on the open-source Unfold/Unfold.jl toolbox (www.unfoldtoolbox.org) and present some new simulation results.