Neurally informed theories on visual working memory

Time/Room: Friday, May 15, 2015, 2:30 – 4:30 pm, Talk Room 1
Organizer(s): Ilja G. Sligte; University of Amsterdam
Presenters: Christian N.L. Olivers, Mark G. Stokes, Ilja G. Sligte, Fiona McNab, Pieter R. Roelfsema, Thomas B. Christophel

< Back to 2015 Symposia

Symposium Description

THEORETICAL IMPASSE How much information can people maintain in working memory and how precise is information in working memory represented? These important questions have produced a fierce tug-of-war between camps of scholars trying to explain capacity limits in working memory in terms of slots or resources. However, that academic debate has nothing to do with how the brain enables working memory. This symposium centers on two central questions that are neurally inspired: Are all working memory representations created equally (working memory states), and how does the neural architecture shape working memory content? WORKING MEMORY STATES Visual working memory has a strict capacity limit of 3-4 items. But do all working memory representations have equal status? Some information might be represented vividly in the center of mind (foreground processes), while other information readily available, but more in the back of the mind (background processes) and yet other information needs an effortful redirection of attention before it is available for report (fragile processes). Christian OLIVERS will provide evidence that when one item is held in working memory, it initially guides attention on a visual search task (as foreground process), but it shifts to the background within a few seconds and is then unable to guide attention. Mark STOKES will show that working memory items are initially actively maintained as persistent activity in monkey prefrontal cortex. However, neural activity soon goes back to baseline, but the information is still represented in the background, presumably in newly configured synaptic weights. Ilja SLIGTE will present evidence that fragile items, that they are normally not available for report, are swapped with background information in working memory when attention is redirected during memory retention. WORKING MEMORY CONTENT Fiona MCNAB will show that the fidelity of working memory representations is impaired when visual information is presented too close together. Apparently, the receptive field size of visual neurons plays a crucial role in constraining the quality of working memory. Pieter ROELFSEMA will present monkey V1 data showing that the contents of working memory are actively maintained in different layers of primary visual cortex by means of top-down projections. Importantly, visual masks transiently erase working memory contents in V1, only to resurface again, presumably through top-down reactivation. Thomas CHRISTOPHEL will show that working memory content can be represented at many different levels in the neural hierarchy, depending on the characteristics of the memoranda. However, the lateral prefrontal cortex cares only about cognitive control, not about working memory content. TARGET AUDIENCE The symposium aims to shift the current debate on working memory to the question how the brain structures and organizes working memory content. We believe this symposium will be interested to students, postdocs, and faculty. The contents and methods will be useful to a large VSS audience: anyone studying working memory or attention, and anyone with interests in multivariate pattern analyses and multilayer electrophysiological recordings. The symposium could benefit them by suggesting new theoretical frameworks to think about data, as well as new experimental methods and paradigms.


On the role of working memory in visual attention

Speaker: Christian N.L. Olivers; VU University Amsterdam

Current cognitive and neural models of visual attention emphasize the role of working memory in biasing attention to task-relevant input. According to these models, the mnemonic maintenance of visual representations automatically creates an attentional template that prioritizes corresponding stimuli for selection. However, the past decade has provided evidence that visual working memory per se is not sufficient, nor necessary for guiding attention. I give a brief review of the field and of behavioral evidence from our lab, using paradigms that combine a memory task with a visual search task. This evidence suggests that for working memory representations to bias visual attention they require a special active template (or ‘foreground’) status – in line with models of working memory that assume an internal focus of attention (Oberauer & Hein, 2012). This while more passive ‘accessory’ or ‘background’ memories do not bias attention. Moreover, our most recent behavioral, eye tracking and EEG experiments indicate that task-relevant representations are actively maintained in working memory for only the first one or two trials, after which the memory representation appears to adopt a less active background status (interpreted as a shift to long term memory; Carlisle, Arita, Pardo, & Woodman, 2011). Intriguingly though, this shift from working memory occurs regardless of whether the memory is being used for attentional guidance or not, thus pointing towards a potential dissociation between active (foreground) vs. passive (background) on the one hand, and biasing attention vs. not biasing attention on the other.

Dynamic Coding for Working Memory in Prefrontal Cortex

Speaker: Mark G. Stokes; University of Oxford

It is often assumed that maintenance in visual working memory is directly supported by persistent activation of the corresponding neural representation. However, the empirical evidence is also quite mixed – persistent delay activity is not always associated with successful memory performance. We propose an alternative to standard ‘persistent activity’ models: working memory can be maintained via ‘activity silent’ neural states, such as temporary changes in effective connectivity. Within this dynamic coding framework, working memory is manifest in a temporary shift in the response profile of a neural circuit. Although such changes in connectivity may be difficult to detect using standard recording approaches, hidden states can be inferred indirectly from changes in network behavior. Here we describe a series of novel multivariate analyses that track population-level dynamics in monkey prefrontal cortex during working memory encoding, maintenance and retrieval. The presentation of a memory item triggers a complex trajectory through activity state space during the initial encoding period that strongly differentiates between memory items. Mean activity levels return to baseline during the maintenance period, however spontaneous spiking patterns continue to reflect the contents of working memory. Finally, the presentation of a memory probe triggers a state-dependent response profile that can be read out for successful memory performance. These characteristics of dynamic coding are consistent with activity-driven changes in underlying connectivity, such as short-term synaptic plasticity and/or network coherence. We consider how such a coding scheme for visual working memory could generalize to other forms of context-dependent behavior.

Multiple levels in visual short-term memory

Speaker: Ilja G. Sligte; University of Amsterdam
Authors: Dirk van Moorselaar1, Christian Olivers1, Victor A.F. Lamme2, Kimron L. Shapiro3; 1VU University Amsterdam; 2University of Amsterdam, 3University of Birmingham

As we go up the visual hierarchy, receptive field size becomes larger, tuning characteristics become more complex, and the lifetime of neural responses increases. As a logical consequence, one would predict increasingly strict capacity limits, loss of visual detail, and longer representational lifetime for representations that depend on higher visual brain regions. Thus, the neural system acts as a low-pass filter limiting capacity, yet increasing lifetime at the same. In this talk, we will provide evidence that the characteristics of visual sensory memory cohere to this principle: our brief and super-capacity iconic memory depends on neural excitability in primary and secondary visual cortex, while our seconds-lasting and high-capacity fragile memory depends on neural activation in higher visual areas. In that sense, iconic memory and fragile memory are just like low-order and high-order forms of visual sensory memory. In addition, we will show that when information from sensory memory is made available for report, it replaces virtually all information that is currently stored in visual working memory, except for one item that remains untouched. Based on the fact that replaced working memory content can be pulled back for report, we argue that there are at least three fundamentally discernable levels in visual short-term memory; 1) foreground processes that form the center of mind, 2) background processes that are readily available for report, but can easily be swapped with 3) fragile, sensory memory representation that passively decay when there is no top-down amplification available.

Competitive interactions affect working memory precision

Speaker: Fiona McNab; University of Birmingham
Authors: Jumana Ahmad1, Anna C. Nobre2, Kimron L. Shapiro1; 1University of Birmingham, 2University of Oxford

Competition between visually presented stimuli is associated with reduced neural firing, longer reaction time and reduced BOLD response. It is not known whether the effects of competition extend to working memory (WM), nor whether competition represents a common limiting-factor, compromising both WM performance in the absence of distraction, as well as effective distractor exclusion. Here we measured WM precision for an item placed with small or large spatial separation from another item to be remembered, or from a distractor. In both cases, WM precision was significantly reduced for small relative to large spatial separation. This indicates that the effects of competition extend to WM precision, and identifies competition as a potential common mechanism affecting WM regardless of whether the two items being encoded are both to be remembered, or whether one is a distractor. Such a mechanism is a potential basis for the association identified between distractor exclusion and WM in the absence of distraction.

The role of the different layers of primary visual cortex in working memory

Speaker: Pieter R. Roelfsema; Netherlands Institute for Neuroscience
Authors: Matthew W. Self, Timo van Kerkoerle; Netherlands Institute for Neuroscience

Imaging studies have revealed a neuronal correlate of working memory in primary visual cortex (Harrison & Tong, Nature, 2009). However, it is unknown if working memories influence spiking activity in the primary visual cortex. To address this question, we recorded neuronal activity in the primary visual cortex of monkeys trained to perform attentional and working memory tasks with a probe that records activity in all the cortical layers. We found a consistent working memory trace in the spiking activity in the superficial and deep layers of monkey V1, and only a weak memory representation in input layer 4. This V1 memory trace could be disrupted with a visual mask, but it then quickly recovered. The advantage of the laminar probe is that is also gives insight into the current-source density, which reveals the putative synaptic sources of memory activity. The current-source density measurements revealed a characteristic signature of feedback processing with putative synaptic inputs in the superficial and deep layers for working memory. This signature resembles the signature of selective attention, supporting the view that top-down modulation of activity in primary visual cortex underlies both working memory and attention. Our results provide new insights into the role of early visual cortex in working memory.

Distributed Visual Working Memory Stores Revealed by Multivariate Pattern Analyses

Speaker: Thomas B. Christophel; Charité Universitätsmedizin
Authors: Chang Yan1, Carsten Allefeld1, John-Dylan Haynes1,2; 1Charité Universitätsmedizin, 2Humboldt Universität zu Berlin

Distributed Visual Working Memory Stores Revealed by Multivariate Pattern Analyses Thomas B. Christophel, Chang Yan, Carsten Allefeld & John-Dylan Haynes The storage buffers retaining visual working memory contents were originally postulated to reside in prefrontal cortex. Recently, a dissenting view has evolved claiming that working memory content depends on distributed storage in sensory brain regions. We provide strong evidence for this claim in a series of fMRI experiments investigating the content-specificity of delay-period activity using multivariate pattern analyses. Representations of color and motion patterns as well as complex shapes were identified in early visual, and lateral occipital posterior parietal cortex, but also in the frontal eye fields. A meta-analysis of content-specificity within these brain areas revealed large inter-areal differences critically depending on whether the stimuli were smooth global patterns or shapes with clear edges and on whether stimuli varied across color, luminance or motion direction dimensions. In addition, we show that areas beyond early visual cortex retain information in an inherently view-independent format and that coding of a given stimulus in higher visual areas is not solely driven by the visual display originally shown. Instead, the representation changes when a subject mentally transforms what they are holding in mind (i.e. during mental rotation). Extending our findings on visual working memory, we show that verbal content (Chinese Characters memorized by native speakers of Chinese) is selectively stored in prefrontal areas, more specifically Broca’s area and articulatory premotor cortex. Finally, while working memory storage seems to be represented in a distributed way, working memory control could be traced to dorsolateral prefrontal cortex regardless of what content was memorized.

< Back to 2015 Symposia