How are humans capable of maintaining detailed representations of visual items in memory? When required to make fine discriminations, we sometimes implicitly differentiate memory representations away from each other to reduce inter-item confusion. However, this separation of representations can inadvertently lead memories to be recalled as biased away from other memory items, a phenomenon termed repulsion bias. Using a non-retinotopically specific working memory paradigm, we found stronger repulsion bias with longer working memory delays, but only when items were actively maintained. These results suggest that (1) repulsion bias can reflect a mnemonic phenomenon, distinct from perceptually driven observations of repulsion bias, and (2) mnemonic repulsion bias is ongoing during maintenance and dependent on attention to internally maintained memory items. These results support theories of working memory where items are represented interdependently and further reveals contexts where stronger attention to working memory items during maintenance increases repulsion bias between them.

No abstract

Visual attention plays an essential role in selecting task-relevant and ignoring task-irrelevant information, for both object features and their locations. In the real world, multiple objects with multiple features are often simultaneously present in a scene. When spatial attention selects an object, how are the task-relevant and task-irrelevant features represented in the brain? Previous literature has shown conflicting results on whether and how irrelevant features are represented in visual cortex. In an fMRI task, we used a modified inverted encoding model (IEM, e.g., Sprague & Serences, 2015) to test whether we can reconstruct the task-relevant and task-irrelevant features of spatially attended objects in a multi- feature (color + orientation), multi-item display. Subjects were briefly shown an array of three colored, oriented gratings. Subjects were instructed as to which feature (color or orientation) was relevant before each block, and on each trial were asked to report the task-relevant feature of the object that appeared at a spatially pre-cued location, using a continuous color or orientation wheel. By applying the IEM, we achieved reliable feature reconstructions for the task-relevant features of the attended object from visual ROIs (V1 and V4v) and Intraparietal sulcus. Preliminary searchlight analyses showed that task-irrelevant features of attended objects could be reconstructed from activity in some intraparietal areas, but the reconstructions were much weaker and less reliable compared with task-relevant features. These results suggest that both relevant and irrelevant features may be represented in visual and parietal cortex but in different forms. Our method provides potential tools to noninvasively measure unattended feature representations and probe the extent to which spatial attention acts as a "glue" to bind task-relevant and task-irrelevant features.

Inverted encoding models have recently become popular as a method for decoding stimuli and investigating neural representations. Here we present a novel modification to inverted encoding models that improves the flexibility and interpretability of stimulus reconstructions, addresses some key issues inherent in the standard inverted encoding model procedure, and provides trial-by-trial stimulus predictions and goodness-of-fit estimates. The standard inverted encoding model approach estimates channel responses (or reconstructions), which are averaged and aligned across trials and then typically evaluated using a metric such as slope, amplitude, etc.). We discuss how this standard procedure can produce spurious results and other interpretation issues. Our modifications are not susceptible to these methodological issues and are further advantageous due to our decoding metric taking into account the choice of population-level tuning functions and employing a prediction error-based metric directly comparable across experiments. Our modifications also allow researchers to obtain trial-by-trial confidence estimates independent of prediction error which can be used to threshold reconstructions and increase statistical power. We validate and demonstrate the improved utility of our modified inverted encoding model procedure across three real fMRI datasets, and additionally offer a Python package for easy implementation of our approach.