EEG-based decoding of shapes and their categories in visual working memory

Visual working memory (VWM) allows us to store information in a highly accessible format for an upcoming task. Traditionally, VWM studies require participants to keep a precise copy of a stimulus in mind. But in the real world, we might need to store the same information for different types of tasks, such as recognition or categorisation judgements. For example, when deciding if a bike is the exact model you want, or the same brand. In this study, we asked how categorisation modulates VWM representations. Participants first learned to group unfamiliar shapes from the Validated Circular Shape (VCS) space (Li et al., 2020) into two categories based on their visual features. They then completed a shape VWM task that either required delayed match-to-sample or delayed match-to-category judgements on different blocks while we collected electroencephalography (EEG) data. We tracked the emergence of stimulus-, category- and task level information with high temporal resolution using multivariate pattern analyses of EEG. The neural activity pattern over posterior electrodes contained information about the memorised shape for about one second following VWM encoding. Initially, the stimulus code overlapped across the two tasks, but quickly separated according to task. Later in the delay, stimulus coding persisted only for the match-to-category task and was accompanied by a neural category signal, indicating that categorisation may require an active stimulus representation. To our knowledge, this is the first illustration that the VCS space is decodable from EEG, preserving the circular similarity structure. This provides a fruitful avenue for researchers looking to characterise neural representations of unfamiliar visual stimuli with high temporal resolution. The results of this study will help elucidate the neural mechanisms supporting VWM under different task demands.