For a long time, working memory was considered to reflect sustained activity in prefrontal and parietal areas. This has been challenged by the "sensory recruitment model" according to which sensory areas are involved in maintaining working memory representations. The main support for this comes from the finding that working memory contents can be decoded from early sensory areas during memory delay periods. However, the availability of stimulus-related information in sensory regions does not reveal whether this information is encoded in a sensory format, i.e. in the same way as sensory stimuli. Alternatively, working memory contents could be encoded in independent formats or even potentially undergo dynamic changes throughout the delay. Here, we test this question directly by requiring participants to briefly memorise naturalistic object-scenes. We directly compared the way early visual regions encoded the contents during perception and working memory maintenance. As in previous studies, we found robust stimulus-related information in early visual cortex throughout the delay. However, despite continuously robust information, the encoding of memory contents in the later delay period did not occur in the same format as during pure perception. Thus, despite the fact that sensory areas have working-memory-related information, the format of this information is not strictly sensory-like throughout the delay.
Magnification-Aware Distillation (MAD): A Self-Supervised Framework for Unified Representation Learning in Gigapixel Whole-Slide Images
arXiv:2512.14796v1 Announce Type: cross Abstract: Whole-slide images (WSIs) contain tissue information distributed across multiple magnification levels, yet most self-supervised methods treat these scales as independent

