Presentation Information
[P3-09]Neural Oscillatory Entrainment in Non-Deterministic Continuous Environments, decoupled from Bayesian Interval Learning
*Elmira Hosseini1,2, Assaf Breska1 (1. Max-Planck Institute for Biological Cybernetics (Germany), 2. Tübingen University (Germany))
Keywords:
Temporal Prediciton,Neural Mechanisms,Non-Deterministic Environments,Computational Modelling,EEG
Predicting the timing of events in continuous, dynamic environments is essential for efficient interaction. In deterministic contexts this is putatively mediated by Oscillatory Entrainment (OE) to the rhythm, and reflected neurally in low-frequency phase alignment, buildup of ramping activity before target, and modulation of target-evoked responses. However, real-world contexts often lack deterministic regularities (e.g., speech). It remains unclear whether and when OE mechanisms engage in non-deterministic continuous streams, and if they can operate separately from distributional learning (DL) processes previously found in uncertain isolated interval conditions. Here, we combined computational modeling of OE (using a simple harmonic coupled oscillator) and DL (using an ideal Bayesian observer) with human EEG recording. We created continuous streams with low (25%) or high (50%) variability, which led to distinct predicted timepoints from the two models. Participants completed a speeded response task with targets at predicted timepoints for each model, as well as intermediate and late timepoints to control for hazard effects. Behaviorally, reaction times were reduced in the 25% relative to 50% condition, selectively for the OE-aligned targets, despite pronounced hazard effect on response times. Neurally, OE-aligned targets elicited lower P3 amplitudes in the 25% relative to 50% condition or to DL-aligned targets, indicating less need for updating for OE predictions. Additionally, delta-band inter-trial phase coherence (ITPC) was higher in the 25% condition before OE target time, mirroring observations in isochronous streams. Interestingly, no contingent negative variation (CNV) was observed. These results highlight the role of oscillatory phase alignment as a predictive mechanism even in the absence of explicit preparatory signals and support the selective engagement of OE in non-deterministic contexts with lower variability, while decoupled from Bayesian DL.