Visuospatial Working Memory Is Cortically Enabled Through Veridical, Categorical and Semantic Representations

09:00 - 10:30

Mechanisms of Short-Term Memory: From Sensory Recruitment to High-Level Abstraction

Room: HSZ - N9

Chair/s:

Thomas B Christophel, Rosanne Rademaker

We are capable of retaining a large variety of visual content in working memory, ranging from simple low-level features to complex naturalistic stimuli. Maintenance of visual information in working memory is accompanied by memory-specific activity across the entire cortical sheet from early visual areas to frontal cortex. Interference between memory content and distractors shapes these representations, as does the passage of time. Here, we bring together experts from cognitive psychology and neuroscience trying to understand the cortical and cognitive mechanisms of short-term memory. Using behavioral work, TMS, single-cell recordings, fMRI, and convolutional neural networks, they assess the representational nature of working memory storage and its interaction with the environment.
First, Pablo Grassi asks whether activity in sensory cortex is necessary for the maintenance of visual information. He will present results from three experiments investigating whether TMS pulses applied over visual cortex interfere with working memory performance for low-level features. Michael Wolff will then show that V1 neurons reverse preference between the processing and short-term maintenance of natural images, evident in both spontaneous and evoked (“pinged”) spiking activity. This suggests that neural adaptation acts like a short-term memory buffer in the early sensory cortex. Next, Anna Zier asks which brain regions represent how low-level visual features (like color and motion) are bound into a more complex object in working memory. Using fMRI decoding, she demonstrates that trial-specific binding information can be identified from memory-related activity in early visual cortex (V1–V4). Then, Anastasia Kiyonaga uses CNN derived similarity measures in natural images to show that low-level and high-level interference uniquely affect working memory performance. Intriguingly, interference effects during working memory are inversely related to long term memory recollection, suggesting competition with immediate memory can strengthen longer-term memory. Finally, Joana Seabra shows that during visual working memory several cortical regions utilize categorical, semantic, and spatial representational formats to maintain simple low-level stimuli in a robust fashion.

Submission 174

Visuospatial Working Memory Is Cortically Enabled Through Veridical, Categorical and Semantic Representations

SymposiumTalk-05

Presented by: Joana Pereira Seabra

Joana Pereira Seabra ^{1, 2}, Andreea-Maria Gui ^{1, 2}, Vivien Chopurian ^{1, 2}, Alessandra S Souza ³, Carsten Allefeld ⁴, Thomas Christophel ^{1, 2}

¹ Department of Psychology, Humboldt-University, Berlin, Germany

² Bernstein Center for Computational Neuroscience Berlin, Germany

³ Faculty of Psychology and Education Sciences, University of Porto, Portugal

⁴ Department of Psychology and Neuroscience, City St George’s, University of London, United Kingdom

Visual stimuli are subject to categorical biases that impact their recall during working memory. Are these systematic errors also evident in the brain? A single visual stimulus might elicit multiple representations throughout the cortex that differ in content and format, with the latter ranging from continuous to categorical. These categorical neural representations likely constitute a neural correlate of the categorical biases evident in behavior.

Forty participants performed an orientation working memory task while fMRI data were recorded. The nature of cortical orientation representations in response to this task was assessed using multivariate encoding modelling. Additionally, we cross-decoded between orientation and congruent verbal and location stimuli. Our results show that several representational formats are present concurrently across the cortex. These are organized along a gradient of abstraction, with more veridical representations in posterior areas, and more abstract, categorical codes in anterior areas. Regarding the semantic and location stimuli, cross-decoding analyses suggest there is a partially shared neural code between orientations and each of these stimuli. Together, these results suggest that neural representations in response to visual stimuli can simultaneously include categorical, semantic, and spatial formats.

Our findings show that visual working memory relies on representational formats ranging from continuous to categorical, including visuospatial simplifications and semantic depictions of the original stimulus.

Bookmark