Ilker Duymaz¹, Micha Engeser^1,2, Daniel Kaiser^1,2,3,4; ¹Justus Liebig University Giessen, Germany, ²Center for Mind, Brain and Behavior (CMBB), Universities of Giessen, Marburg and Darmstadt, ³Center for Applied Computer Science and Data Science (ZAD), Justus Liebig University Giessen, Marburg and Darmstadt, ⁴Cluster of Excellence “The Adaptive Mind”, Universities of Giessen, Marburg and Darmstadt

Multivariate analyses of M/EEG data commonly rely on aligning neural signals to stimulus onsets. Such approaches typically show high decoding accuracy during the transient onset response (within the first 500 ms), which then drops to a lower, more sustained level. In this study, we investigated time-resolved EEG decoding for natural scene processing under conditions where scenes gradually enter the visual field without a discrete onset. To this end, we generated video sequences in which one scene category (e.g., a beach) smoothly transformed into a different category (e.g., a forest). This was implemented by blending images from two categories into a composite and moving a square aperture across it so that the visible content continuously shifted from one scene to the other. When decoding the category of the first and second scene in these videos from the EEG signal, we observed a striking contrast between decoding for the first and second scene. For the first scene, we found robust decoding peaking around 70 ms after onset with a discernible peak structure. In contrast, the second scene category yielded much lower decoding with no clear peak structure. Aligning the emergence of category-diagnostic information for the second scene using deep neural network features did not improve decoding or recover a peak structure. Classifiers trained on the first scene generalized to the second, but with a broad and diffuse temporal pattern, suggesting the second scene did not follow the hierarchical temporal cascade evident for the first scene. These results show that sudden versus gradual onsets produce distinct temporal dynamics, implying that results from onset-based decoding studies do not readily generalize to free-flowing natural stimuli.

Acknowledgements: This work was supported by the European Research Council (ERC) Starting Grant (PEP, ERC-2022-STG-101076057), under Germany's Excellence Strategy (EXC 3066/1, "The Adaptive Mind", project no. 533717223), and by the Deutsche Forschungsgemeinschaft (DFG) grants KA4683/5-1 and KA4683/6-1.