Janna Wennberg¹, Kirsten Adam², John Serences¹; ¹University of California, San Diego, ²Rice University

Working memory (WM) is surprisingly robust to sensory distraction, but there is debate about how neural codes protect WM against sensory interference (Kamitani & Tong, 2005; LaRocque et al., 2012; Rademaker et al., 2019). Some studies propose that memory representations transform within early visual areas to minimize interference (Libby & Buschman, 2021), while others suggest that these memory representations are offloaded to frontoparietal regions, reducing early visual regions’ role in WM storage (Xu, 2021). We used fMRI to test whether early visual regions “multi-task” to concurrently support WM and the processing of incoming sensory information. Eight participants performed a spatial WM task, remembering the angular spatial position of a flickering checkerboard circle. On half of the trials, the screen was blank during the delay; on the other half, identical checkerboard circles flickered in other locations. We used circular ridge regression to decode spatial position, testing within-condition and cross-condition classification accuracy. The frontoparietal account predicts that sensory distractors degrade WM representations, leading to poor classification accuracy in early visual regions when training and testing on distractor-present trials. The flexible multi-tasking account predicts a cross-over interaction because mnemonic information is transformed but not lost. Decoding accuracy was similar in the distractor-present and distractor-absent conditions, consistent with the lack of a behavioral effect on recall error. However, we observed a crossover interaction as early as V1: Training on distractor-absent trials yielded better decoding for distractor-absent trials than distractor-present trials, and vice-versa. Cross-time generalization analyses showed a particularly robust cross-over interaction when training on late-delay period data. Notably, this interaction was absent in areas such as IPS. While our findings accommodate a prominent role for frontoparietal regions in storage and maintenance, they also suggest that voxel population codes in early visual regions flexibly transform to support storage of mnemonic information.

Acknowledgements: This work was supported by RO1-EY025872 awarded to John T. Serences