Recent findings suggest that posterior parietal cortex (PPC) represents information retrieved from long-term and short-term memory. However, the nature and quality of parietal memory representations remain largely unknown. Here, we tested whether exemplar-level details of perceived and remembered stimuli are represented in PPC, using a recently developed method that allows for individual face images to be reconstructed from fMRI activity patterns (Cowen, Chun, & Kuhl, 2014). The experiment consisted of two phases: perception and working memory. During the perception phase, participants viewed hundreds of faces while performing a continuous recognition task, where they judged whether each image was repeated within a block or not. For the working memory phase, we employed a retro-cue paradigm (Harrison & Tong, 2009) in which two faces were presented in rapid succession followed by a cue to maintain one of the two faces. We first tested whether individual faces can be reconstructed from PPC activity patterns elicited during perception. We estimated a regression model that mapped face components to multi-voxel activity patterns using a set of training faces. Reconstructions were then generated for a distinct set of test faces by taking linear combinations of the predicted component weights. Reconstructions created from PPC were more similar to the actually viewed face than other test faces, indicating that PPC distinguishes individual face images. We further explored whether we could reconstruct faces retrieved from working memory. We trained the model on the perception phase data and applied it to the patterns obtained during the working memory delay period. Reconstructions generated from the delay period activity in PPC exhibited above-chance similarity to the original face, suggesting that the contents of memory can be reconstructed from PPC activity patterns. Together, these findings indicate that PPC activity patterns reflect exemplar-level details of visual stimuli during perception and retrieval from memory.
展开▼
