Are task-irrelevant object features stored in working memory in a hidden state?

Existing evidence suggests that a single, task-relevant feature of a multi-feature object can be selectively stored in working memory (WM). Specifically, studies have shown that task-irrelevant object properties receive less priority during encoding and are not decodable from neural activity recorded throughout the delay period. Prominent WM theories posit that items in WM can be stored in different states of activation depending on the allocation of attention such that unattended-but-stored information can be represented in sub-threshold neural activity that is not directly observable using methods such as multi-variate pattern analysis (MVPA). However, it is possible to momentarily “reawaken” these representations using single-pulse TMS (Rose et al., 2017) or by flashing a task-irrelevant stimulus during the delay (Wolfe et al., 2017). In the present study, we used similar methods to examine whether task-irrelevant features of a single object are stored in WM in a similar hidden state. Specifically, we used MVPA trained on EEG data to examine the temporal evolution of the task-relevant and irrelevant feature representations of a two-feature object. In different blocks, participants remembered the orientation or color of a colored, oriented grating, and, during the delay period, a task-irrelevant stimulus was flashed on the screen. A support vector machine classifier was then used to classify the identity of the stimulus features. Our results replicated previous findings showing that orientation was only decodable throughout the delay period when it was task-relevant. However, analysis of the signals evoked by the presentation of the flash stimulus revealed above-chance decoding for both task-relevant and irrelevant orientations, although decoding accuracy was generally higher and less variable across subjects for task-relevant versus irrelevant orientations. These results suggest that task-irrelevant features may be stored in WM in a hidden state and that participants may vary in their ability to selectively store only task-relevant features.