HHS Public Access
Author manuscript
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Published in final edited form as:
Trends Cogn Sci. 2013 March ; 17(3): 134–141. doi:10.1016/j.tics.2013.01.010.
Flexible cognitive resources: competitive content maps for 
attention and memory
Steven L. Franconeri1, George A. Alvarez2, and Patrick Cavanagh2,3
1Department of Psychology, Northwestern University, Chicago, USA
2Vision Sciences Laboratory, Harvard University, Cambridge, USA
3Laboratoire Psychologie de la Perception, Université Paris Descartes, Paris, France
Abstract
The brain has finite processing resources so that, as tasks become harder, performance degrades. 
Where do the limits on these resources come from? We focus on a variety of capacity-limited 
buffers related to attention, recognition, and memory that we claim have a two-dimensional ‘map’ 
architecture, where individual items compete for cortical real estate. This competitive format leads 
to capacity limits that are flexible, set by the nature of the content and their locations within an 
anatomically delimited space. We contrast this format with the standard ‘slot’ architecture and its 
fixed capacity. Using visual spatial attention and visual short-term memory as case studies, we 
suggest that competitive maps are a concrete and plausible architecture that limits cognitive 
capacity across many domains.
Understanding cognitive capacity limitations
When observers are asked to deal with too much information, too many tasks, or too many 
targets processing becomes slower or less accurate. Why? A typical answer is that the brain 
has only a finite ‘capacity’ for processing and, because these limited ‘resources’ are spread 
more thinly with increased task ‘load’, speed and accuracy must be sacrificed (e.g., [1,2,3]). 
Unfortunately, words such as ‘capacity’, ‘resources’, and ‘load’ relabel the effect without 
explaining why it occurs. Despite this circularity, the concept of limited resources has 
become central to cognitive research. So, what is the resource? Where does it reside and why 
is it limited? Can people get more of it?
There are a number of possible resources and we focus here on the temporary buffers that 
hold information for analysis and control, specifically buffers for attention and working 
memory, whose capacity will directly determines the complexity and quantity of processing 
that individuals can manage. Map representations (Box 1), found widely throughout many 
brain areas [4], are a likely format for these buffers. We suggest that competitive interactions 
Corresponding author: Franconeri, S.L. (franconeri@northwestern.edu). 
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our 
customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of 
the resulting proof before it is published in its final citable form. Please note that during the production process errors may be 
discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 2
(e.g., [5,6]) between items within these map representations provide a direct explanation of 
the capacity limits in cognition. In this two-dimensional ‘map’ architecture, individual items 
must compete for actual, bounded space. This architecture defines a flexible resource that is 
physical rather than metaphorical: it is cortical real estate. Our goal is to describe this 
identifiable, measurable, and accessible ‘space’ as a concrete explanation for phenomena 
such as load and capacity. We will contrast this map architecture, with its focus on locations 
and competition within a limited space, with ‘slot’ architectures, which have a fixed number 
of places in which information can be stored.
Box 1
Map representations
In the brain, a region can be considered a ‘map’ if it has a coherent spatial organization 
where the preferred stimuli of neurons change smoothly from one location to the adjacent 
one. Because of the layering of neural architecture, cortical maps are constrained to have 
one or two dimensions of global organization, either continuous dimensions, such as 
space for the FEF (a) or less systematic clusters (e.g., patches of related shape identities). 
Many maps also have local embedded dimensions. The retinotopic visual field map of 
area V1 (b) is globally spatially organized, with embedded dimensions including 
orientation, eye of origin, spatial frequency, and color (see [4]). A motion map (MT) 
differs in the embedded dimensions (motion direction, direction polarity; see [66]). A 
tonotopic frequency map of incoming auditory information is represented in primary 
auditory cortex (Heschl’s Gyrus), organized by frequency ([67]), with evidence for 
inhibitory surrounds within that space ([68]).
Some maps do not show obvious continuous dimensions but do show local clusters of 
related values. For example, shape maps (c) are proposed in area TE of the ventral visual 
stream (image adapted from [69]), representing as yet unknown feature dimensions, 
including progressive transitions in face space [70]. There may be similar ‘clustered’ 
maps for other types of representations, including phonemes (posterior superior temporal 
gyrus [57]) and principal components of odor space [71,72].
Other maps represent action plans and goals, serving not as a sensory representation but 
as ‘source code’ that underlies behavior [73]. One example is the explicitly spatial eye 
movement maps described in Figure 2a. There is also evidence for ‘clustered’ motor plan 
maps (d). In monkeys, precentral motor representations that are coarsely somatotopically 
organized (e.g., hand movement areas tend to be near finger movement areas) also 
contain clusters of different motor plans related to the same body area [74] and there is 
also evidence for inhibitory surrounds in that space [75].
Map representations offer a rich set of computational advantages ([73], but see also [76]). 
They allow fast parallel computation with minimized axon length between mutually 
relevant information [73,77]. Map addressing is error-tolerant – getting rough instructions 
for the carrots in a supermarket will at least get you to the vegetable section, in contrast 
to an almost-correct phone number which gets you nowhere. Finally, maps with shared 
coordinate systems benefit from straightforward cross-referencing of information, for 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 3
example, for spatial representations derived from both visual and auditory information in 
the superior colliculus [78].
Box 1. 
Examples of cortical maps
Map vs slot architectures
In a map architecture, each position represents a value within a specific information space, 
such as spatial location, color, or shape. Such two-dimensional maps are widely reported in 
sensory and motor representations of various levels of complexity (Box 1; see [4,7], for 
review). The capacity of a map is flexible, limited by the space taken up by the activity 
profile of individual items on the map, how they interact with each other, and the spacing of 
the items on the map. Items interact destructively when they are close enough for their 
activity profiles to overlap, due to the inhibition zone that typically surrounds each activity 
peak [8]. These suppressive surrounds sharpen the activity profiles of single items and 
resolve inter-item competition – a critical step especially when unitary actions are needed 
(e.g., a saccade to a single location). These competitive interactions mean that map capacity 
is not fixed, but determined both by the number and arrangement of the items within it 
(Figure 1).
Previous proposals suggest that items that are cortically closer are more likely to compete 
with each other for representation (e.g., [5,6,8]). More recent proposals have suggested that 
such competitive interactions are the roots of capacity limits for tasks such as object 
recognition [9] and multiple object tracking [10]. Here, we suggest that these examples of 
inter-item competition combined with the anatomical properties underlying cortical maps 
provide a concrete explanation of the flexible capacity of attentional and short-term memory 
resources, serving as case studies for other cognitive resource limits. Representations related 
to more complex abilities, such as semantic identification or task control, may not be so 
obviously organized as a two-dimensional space (see [11–13] for discussion of alternate 
formats), but we will outline the capacity effects that should result if this were the case.
An alternative to the map format is a slot architecture, which stores information across a 
fixed number of independent locations. The location of a slot is unrelated to its contents, and 
serves only as an address to return to when information must be retrieved. Unlike 
competitive maps, the arrangement of items is irrelevant (Figure 1). Examples include 
characterizations of visual spatial attention (see [14], for review), as well as models of visual 
short-term memory [15,16] (Box 2). The capacity of the slot architecture is set by the 
number of items that can be held (typically one per slot), with an independent information 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 4
limit for each item. Slot models often cite temporal neural synchrony limitations as the root 
of the slot limitation (e.g., [17]), but the item limit is typically set by fiat – there are as many 
slots as the behavioral data warrant.
Box 2
The architecture of visual short-term memory
There is currently considerable debate about whether visual working memory capacity is 
best characterized by fixed slot or flexible resource models. Early research supported a 
slot-model architecture, in which memory was limited only by the number of objects 
stored, independently of the number of features stored per object [79,80]. The strong 
version of this claim has been challenged by a variety of empirical findings [82,82], with 
the current debate centered on the discovery of a tradeoff between the number of items 
stored and the precision of each item [38,83]. Flexible resource models predict this 
tradeoff, but are too flexible to firmly predict how many items can be stored and the 
precision with which they will be stored. Most find their support by fitting the data to a 
continuous function relating number of items to precision (e.g., [83–85]).
Slot models cannot account for the quantity-precision tradeoff without significant 
upgrades. The most straightforward modification is to construct a hybrid model, where 
the number of slots limits the number of objects that can be stored and the amount of 
some other cognitive resource determines the precision with which they can be stored 
[16,86]. Alternatively, the slots themselves can be treated as discrete chunks of resources, 
where memory can store multiple copies of objects in separate slots and then average 
those copies to increase the precision of a subset of item representations [15].
Although these models compete to explain the quantity-precision tradeoff function, 
recent research is converging on a wide range of findings that are not yet addressed by 
either class of model. In particular, working memory representations appear to be 
hierarchical and structured [87,88], contrasting with standard models that focus only on 
individual item capacity (which may be impossible to define even in simple displays). 
Such structured representations may be accommodated by a system in which the contents 
of memory are integrated across multiple competitive maps. An important direction for 
future research is to determine how multiple maps would accommodate these structured 
memory representations and what constraints this would impose.
We suggest that the map architecture is the better alternative because it best explains why 
capacity varies across different kinds of information and different tasks. We illustrate this 
point with two case studies: the capacity of spatial attention and the capacity of visual short-
term memory (VSTM). We then suggest new examples of cognitive limitations that we 
believe could be explained in a similar way.
Case study I: map representations for eye movements and visual selection
A vast field of research has explored the properties of the spatially localized ‘spotlight’ [3] 
of visual attention. These attention hotspots appear to be controlled by the locations of 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 5
target-related activity on two-dimensional maps, such as the frontal eye fields (FEF), which 
were initially thought to be solely for the control of the eyes (see [18,19]). A localized 
current delivered to the retinotopic map triggers an eye movement to the corresponding 
spatial location. However, a weaker stimulation that does not trigger a saccade still affects 
visual processing at the corresponding retinotopic location. For example, stimulating the 
FEF enhances responses at corresponding locations in V4 [20]. Another study [21] showed 
related results for stimulation of the superior colliculus, a subcortical saccade/attention map. 
When not actively directing saccades, these maps, therefore, serve a second function: an 
attention map that acts through downward connections to visual cortex.
Activity peaks in saccade areas also engage large suppressive surrounds that allow stronger 
targets to suppress weaker ones in the competition to be the single, executed saccade. 
Attentional foci also come with suppressive surrounds [22–27] that may be directly related 
to those of the saccade areas. The suppression is beneficial for the selection of the saccade 
target, but also fills a critical role when these areas serve as attention maps, by preventing 
the selection of nearby distractors that would compete for target identification (Figure 2a). 
However, this competitive suppression also limits the number of locations that can be 
simultaneously attended. This is a major component of the resource limit for competitive 
maps, as well as the source of the flexibility in this limit.
Such competition may impose the performance limit on both static multifocal selection tasks 
[10] and dynamic multiple object tracking tasks [10,28–30]. Despite frequent claims that 
such abilities are limited to three–four locations or objects (see [14], for review), recent 
work shows that these limits are malleable in the ways that a map-based explanation predicts 
[31]. Moving the selected locations closer can reduce the limit to one or two, whereas 
moving them farther apart can increase the limit to eight or nine [14,32]. Figure 2b depicts 
how these limits depend on the distribution of items in the visual field [9,33–35], suggesting 
that they arise from competition within lower level visual areas, where each hemifield is 
represented by a separate map [36]. The layout and interactions within a map are likely 
malleable with training. Extensive practice with video games has been shown to improve 
performance on multiple object tracking tasks [37] and this effect may be due to a reduction 
in the spatial extent of the suppressive surround of each item, allowing more objects to be 
tracked.
Case study II: capacity limitations on visual short-term memory
Slots are limited to a set number of independent items. Maps are limited by the types of 
information and its layout, and, thus, predict flexible capacities depending on task 
parameters. This distinction parallels recent debates in the visual memory literature over 
whether capacity is limited by fixed slots or flexible resources (Box 2). Here, we focus on 
the properties of a competitive-content map that can explain flexible memory limits. In 
particular, the degree of inter-item competition on such a map, and, therefore, its carrying 
capacity, will depend on the amount of space required to represent an item on a map, the 
spacing between items, the extent of surround inhibition, and the reduction of suppressive 
surrounds that cross the vertical midline. As an example of the change in overall space taken 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 6
by an item, Figure 3 depicts the tradeoff between the difficulty of recognizing an object’s 
identity and the number that can be stored [38].
Examples of the effects of physical spacing are seen in a number of studies on simultaneous 
visual identification. For example, increased spacing among items within spatially organized 
visual data maps increases the speed and accuracy of visual identification tasks [35,39] and 
visual search tasks [40–42], and may be a key constraint on our ability to identify multiple 
items in parallel [9]. The reduction of inter-item suppression with greater spacing would 
affect both identification and retention in memory.
Brain imaging also suggests that nearby items decrease the representational strength of a 
target item through some form of surround suppression [9,43,44] and this inhibitory effect 
should influence not just visual identification, but also visual memory retention [45]. Such 
inter-item competition during identification can be reduced by separating items across the 
visual hemifield boundary for both identification [35,46] and memory retention [45], which 
suggests again that the hemifield boundary reduces competition across otherwise adjacent 
areas of a map (Figure 1).
One way to reduce this local competition may be to process a single item at a time on a map. 
When items are presented in isolation as a sequence over time, the quality of visual 
representations improves for both recognition [9] and visual memory [45,47–49]. This 
temporal isolation can be mimicked by isolating a subset of objects with selective attention, 
suppressing other items in the map and preventing them from competing. This can improve 
the quality of visual representations in recognition [9,44] and visual memory tasks [50].
Competitive interactions within strictly visual maps are not likely to be the only limiting 
factor on visual memory performance. For example, the function that relates information 
load to capacity shows that, even for items with minimal competition at the recognition stage 
(as measured by visual search performance), the storage limit is still at most approximately 4 
or 5 objects (though this interpretation of this limit is debated, see Box 2). Therefore, in 
addition to competition between the objects within visual maps (e.g. as described in Case 
Study I), there is likely more competition within frontal or parietal structures during later 
memory maintenance [51,52] These parietal and frontal structures may also be organized as 
rough spatial maps, becoming activated primarily when storing information from particular 
spatial positions (see [53], for review; see also [54] for evidence of multiple overlapping 
maps in frontal cortex). Lesions of portions of these frontal maps can even create ‘memory 
scotomas’, where memory representations (but not online perceptual representations) for 
spatial positions are impaired within particular regions of the visual field [55].
We take the map architecture for memory as a plausible assumption: an implementation of a 
cognitive ‘resource’ that has been otherwise vaguely specified.
Maps as limitations for broader cognitive resources
Our case studies of maps within the perceptual system can be extrapolated to maps of 
broader information spaces across cognition, from task scheduling to social reasoning. There 
is strong evidence that many cortical areas are locally specialized for high-level 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 7
representations, such as faces, bodies, or places [56]. Map architectures likely underlie even 
such high-level functions, given the planar structure of the neuronal layers of the cortex (as 
well as many subcortical areas). It is easy to imagine ‘clustered’ maps for many high-level 
spaces and indeed they have been suggested for ‘phonemic maps’ (posterior superior 
temporal gyrus; [57]), specialized ‘letter maps’ in the visual word form area (left fusiform 
gyrus; see [7]), supramodal maps of ‘emotional expression’ with separate subregions for 
representing variations in anger, disgust, fear, happiness, and sadness (MPFC and STS, 
[58]), or even maps of abstract ‘semantic knowledge’, such as vegetables and tools (e.g., 
[59]). And like the ‘source code’ maps for ‘action plans’ organized by body part or action 
goal (Box 1), one might imagine a ‘task buffer’ that stores the type and state of current tasks 
within a clustered space of possibilities. Suppressive surrounds have been demonstrated for 
memory representations in many such high-level spaces, including inhibition of items that 
are semantically [60,61] and orthographically [62] similar to items in memory.
Although such maps are easy to imagine, we focus on what such structures imply about the 
nature of cognitive capacity limitations. Across all of these examples, the carrying capacity 
in each of these spaces would be flexible, determined by the spacing and suppressive 
surrounds of items being represented. Even if these higher-level maps are only roughly 
hinted at by brain imaging and multiunit recordings, a competitive map format implies a 
common set of properties for their capacity limitations. Computational modeling of these 
maps could lead to new insights and predictions about the roots and connections among the 
limitations of each map type [63,64]. As more is learned about the layout, resolution, and 
surround suppression properties of each map (see Box 4 for examples of open questions), 
including how such properties change with experience [65], a competitive map account 
predicts how cognitive capacity should vary based on moment-to-moment content.
Box 4
Outstanding questions
• Do some maps act as ‘pointers’ to other maps? Attentional pointers 
may index the features of a target by specifying its coordinates [19]. A 
task planning map may need to cross reference maps of object features 
to specify the task target, perhaps by specifying the color of the object 
to be picked up.
• Could variation in surround suppression be used to change the 
‘computation’ occurring on a map? For example, if surround 
suppression is disabled, individual items lose their isolated peaks and 
activity is aggregated across several items. This ‘ensemble’ mode could 
provide a substrate for summary representations of a space, such as the 
histograms required to generate perception of featural ‘averages’ of 
dimensions, such as size, orientation, location, or higher-level 
identities, such as facial emotions (see [96], for review).
• What would be the consequence of local competition in ‘clustered’ 
information spaces with less systematic dimensions of organization? 
Would arbitrary clustering lead to idiosyncratic capacity constraints and 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 8
idiosyncratic inter-item interference? For example, on somatosensory 
maps, the hand representation is adjacent to the face and the genitals 
next to the feet, resulting in unusual transfer of sensation for 
individuals with phantom limbs (see [97]). Would there be similar 
idiosyncratic adjacencies and interference patterns for semantic or 
shape maps?
• Can studies exploring maps of functional selectivity (e.g., physiological 
recordings, functional MRI) reveal consistent and stable map 
architectures within individuals for more abstract information spaces 
(e.g., executive planning, semantic processing, social reasoning) in a 
manner akin to mapping the large scale organization in the visual 
stream (e.g., [98]), leading to specific predictions regarding processing 
limits for more abstract cognitive tasks?
Concluding remarks
The human brain depends on a variety of temporary buffers to retain information of current 
interest. We propose that the commonly found map organizations seen throughout the brain 
can give a concrete explanation of their capacity limits.
Maps are 2D spaces of potential sensory and motor representations, such as spatial location, 
visual features, or motor plans. Unlike slot architectures, where capacity limits are fit to the 
data, the 2D representations proposed here predict that capacity limits are flexible, 
constrained by competition for space within the bounded size of each map. Competition 
decreases with distance and sufficient distance (or an anatomical boundary) eliminates 
competition, creating an ‘independent cognitive resource’. We propose that capacity limits 
across the cognitive system may be best understood as competition within 2D maps, from 
attention and memory to motor control and executive planning.
The bottleneck in determining the roots of cognitive capacity limits is not a lack of effort, 
but instead a lack of concrete suggestions for how these mental resource limits might arise 
from properties of neural organization of information storage in the brain. We hope that the 
principle of the competition-limited cognitive map will serve as a concrete suggestion to 
guide future research into the nature of cognitive capacity limits.
Acknowledgments
This work was supported by an NSF CAREER Grant BCS-1056730 to S.F. and an NSF CAREER Grant 
BCS-0953730 to G.A.A., and by NIH EY09258 and Chaire d’Excellence grants to P.C. We thank Doug Bemis, 
Brandon Liverence, Audrey Lustig, and Yangqing Xu for comments, and Kevin Hartstein for his assistance in 
manuscript preparation.
Glossary
Binding problem:
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 9
independent processing of different types of information (e.g., color vs motion) requires that 
the information be later unified (e.g., for visual recognition). Typical solutions to this 
problem rely on linking features by spatial or temporal proximity
Dorsal visual stream:
a pathway of visual processing that projects primarily to the parietal lobe, associated with 
spatial attention and the guidance of action
Ensemble processing:
global processing of a set of objects, resulting in abstracted or statistical representations of 
the set
Feature integration theory:
a proposed solution to the binding problem, which suggests that visual features of an object 
(e.g., color, shape, motion) that are processed in spatially segregated maps are integrated 
when attention is directed to the object’s location [88]
Frontal eye fields:
retinotopically organized maps in the primate frontal cortex, involved in the control of eye 
movements and attention
Map architecture:
a model of cognitive resources that proposes two-dimensional representations of globally or 
locally (clustered) continuous information spaces. Map capacity is flexibly determined by 
the number and distribution of activity ‘peaks’ within the space
Multiple object tracking:
paradigm designed by [99] that tests multifocal selection of moving objects
Semantic Identification:
recognition of an object as a specific instance of a general class
Slot architecture:
a model of cognitive resources that proposes a fixed number of ‘slots’ for storing items, with 
an independent information limit on each slot
Somatotopy:
topographic mapping of body areas to spatially correlated locations in a cortical map
Supramodal map:
higher-level cortical map that transcends particular sensory modalities (e.g., vision, tactile, 
audition) of environmental stimuli
Surround inhibition:
the suppressive effect of an activity ‘peak’ on a map on neighboring areas of the map
Temporal neural synchrony:
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 10
proposed temporal solution to the binding problem, in which spatially segregated brain 
regions synchronize activity related to a given item into a specific phase cycle
Ventral visual stream:
set of hierarchical, retinotopically organized maps from the primary visual cortex to the 
temporal cortex, with a strong role in object recognition
Visual search:
a perceptual task in which an observer scans a visual scene for a particular object or feature 
(target) among other objects and features (distractors)
Visual hemifield:
half of a visual scene, usually split vertically into left and right hemifield. This partition 
arises from the branching of the optic nerve at the optic chiasm, such that information from 
the left half of a visual scene is initially processed in the right hemisphere of the brain and 
vice-versa
Visual short-term memory (VSTM):
capacity-limited memory that stores abstracted versions of visual sensory input for several 
seconds
Visual spatial attention
selective amplification of some locations, features, or objects in the visual field
References
1. Broadbent, D. Perception and Communication. Pergamon Press; 1958. 
2. Norman DA, Bobrow DG. On data-limited and resource-limited processes. Cogn Psychol. 1975; 
7:44–64.
3. Posner MI, et al. Attention and the detection of signals. J Exp Psychol Gen. 1980; 109:160–174.
4. Mountcastle VB. The columnar organization of the neocortex. Brain. 1997; 120:701–722. [PubMed: 
9153131] 
5. Desimone R, Duncan J. Neural mechanisms of selectivevisual attention. Annu Rev Neurosci. 1995; 
18:193–222. [PubMed: 7605061] 
6. Reynolds JH, et al. Competitive mechanisms subserve attention in macaque areas V2 and V4. J 
Neurosci. 1999; 19:1736–1753. [PubMed: 10024360] 
7. Dehaene S, Cohen L. Cultural recycling of cortical maps. Neuron. 2007; 56:84–98.
8. Tsotsos JK, et al. Modeling visual attention via selective tuning. Artif Intell. 1995; 78:507–547.
9. Scalf P, Beck D. Competition in Visual Cortex Impedes Attention to Multiple Items. J Neurosci. 
2010; 30:161–169. [PubMed: 20053898] 
10. Franconeri SL, et al. Tracking multiple objects is limited only by object spacing, not speed, time, 
or capacity. Psychol Sci. 2010; 21:920–925. [PubMed: 20534781] 
11. Stewart TC, et al. Neural representations of compositional structures: Representing and 
manipulating vector spaces with spiking neurons. Connect Sci. 2011; 23:145–153.
12. Bowers J. On the biological plausibility of grandmother cells: implications for neural network 
theories in psychology and neuroscience. Psychol Rev. 2009; 116:220–251. [PubMed: 19159155] 
13. O’Reilly RC. Six principles for biologically based computational models of cortical cognition. 
Trends Cogn Sci. 1998; 2:455–462. [PubMed: 21227277] 
14. Franconeri SL, et al. How many locations can you select? J Exp Psychol Hum Percept Perform. 
2007; 33:1003–1012. [PubMed: 17924803] 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 11
15. Zhang W, Luck SJ. Discrete fixed-resolution representations in visual working memory. Nature. 
2008; 453:233–235. [PubMed: 18385672] 
16. Awh E, et al. Visual working memory represents a fixed number of items regardless of complexity. 
Psychol Sci. 2007; 18:622–628. [PubMed: 17614871] 
17. Raffone A, Wolters G. A cortical mechanism for binding in visual working memory. J Cogn 
Neurosci. 2001; 13:766–785. [PubMed: 11564321] 
18. Awh E, et al. Visual and oculomotor selection: links, causes and implications for spatial attention. 
Trends Cogn Sci. 2006; 10:124–130. [PubMed: 16469523] 
19. Cavanagh P, et al. Visual stability based on remapping of attention pointers. Trends Cogn Sci. 
2010; 14:147–153. [PubMed: 20189870] 
20. Moore T, et al. Visuomotor origins of covert spatial attention. Neuron. 2003; 40:671–683. 
[PubMed: 14622573] 
21. Müller JR, et al. Microstimulation of the superior colliculus focuses attention without moving the 
eyes. Proc Natl Acad Sci U S A. 2005; 102:524–529. [PubMed: 15601760] 
22. Bahcall DO, Kowler E. Attentional interference at small spatial separations. Vis Res. 1999; 39:71–
86. [PubMed: 10211397] 
23. Cave KR, Zimmerman JM. Flexibility in spatial attention before and after practice. Psychol Sci. 
1997; 8:399–403.
24. McCarley JS, et al. Spatially mediated capacity limits in attentive visual perception. Acta Psychol. 
2007; 126:98–119.
25. McCarley JS, Mounts JR. On the relationship between flanker interference and localized 
attentional interference. Acta Psychol. 2008; 12:102–109.
26. Mounts JR. Evidence for suppressive mechanisms in attentional selection: Feature singletons 
produce inhibitory surrounds. Attent Percept Psychophys. 2000; 62:969–983.
27. Cutzu F, Tsotsos JK. The selective tuning model of attention: psychophysical evidence for a 
suppressive annulus around an attended item. Vis Res. 2003; 43:205–219. [PubMed: 12536142] 
28. Carlson TA, et al. Quadrantic deficit reveals anatomical constraints on selection. Proc Natl Acad 
Sci U S A. 2007; 104:13496–13500. [PubMed: 17673552] 
29. Franconeri SL, et al. Multiple object tracking is limited by crowding, but not speed. Psych Bull 
Rev. 2008; 15:802–808.
30. Shim WM, et al. Spatial separation between targets constrains maintenance of attention on multiple 
objects. Psych Bull Rev. 2008; 15:390–397.
31. Franconeri, SL. The nature and status of visual resources. In: Reisberg, D., editor. Oxford 
Handbook of Cognitive Psychology. Oxford University Press; (in press)
32. Alvarez GA, Franconeri SL. How many objects can you track?: Evidence for a resource-limited 
tracking mechanism. J Vis. 2007; 7:1–10.
33. Alvarez GA, Cavanagh P. Independent resources for attentional tracking in the left and right visual 
fields. Psychol Sci. 2005; 16:637–643. [PubMed: 16102067] 
34. Alvarez GA, et al. Anatomical constraints on attention: hemifield independence is a signature of 
multifocal spatial selection. J Vis. 2012; 12:1–20.
35. Torralbo A, Beck DM. Perceptual load-induced selection as a result of local competitive 
interactions in visual cortex. Psychol Sci. 2008; 19:1045–1050. [PubMed: 19000216] 
36. Gardner JL, et al. Maps of visual space in human occipital cortex are retinotopic, not spatiotopic. J 
Neurosci. 2008; 28:3988–3999. [PubMed: 18400898] 
37. Green CS, Bevallier D. Action video game modifies visual selective mechanism. Nature. 2003; 
423:534–537. [PubMed: 12774121] 
38. Alvarez GA, Cavanagh P. The capacity of visual short-term memory is set both by visual 
information load and by number of objects. Psychol Sci. 2004; 15:106–111. [PubMed: 14738517] 
39. Balas B, et al. A summary statistic representation in peripheral vision explains visual crowding. J 
Vis. 2009; 9:13. [PubMed: 20053104] 
40. Rosenholtz R, et al. A summary statistic representation in peripheral vision explains visual search. 
J Vis. 2012; 12:14. [PubMed: 22523401] 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 12
41. Fei-Fei L, et al. Why does natural scene categorization require little attention? Exploring 
attentional requirements for natural and synthetic stimuli. Vis Cogn. 2005; 12:893–924.
42. Reddy L, VanRullen R. Spacing affects some but not all visual searches: implications for theories 
of attention and crowding. J Vis. 2007; 7:3. [PubMed: 18217818] 
43. Parks NA, et al. Steady-state signatures of visual perceptual load, multimodal distractor filtering, 
and neural competition. J Cogn Neurosci. 2011; 23:1113–1124. [PubMed: 20146614] 
44. McManis SA, Somers DC. Processing efficiency of divided spatial attention mechanisms in human 
visual cortex. J Neurosci. 2005; 25:9444–9448. [PubMed: 16221854] 
45. Umemoto A, et al. A bilateral advantage for storage in visual working memory. Cognition. 2010; 
117:69–79. [PubMed: 20659731] 
46. Mathot S, et al. From reorienting of attention to biased competition. Attent Percept Psychophys. 
2010; 72:651–657.
47. Ihssen N, et al. Improving visual short-term memory by sequencing the stimulus array. Psych Bull 
Rev. 2010; 17:680–686.
48. Gorgoraptis N, et al. Dynamic updating of working memory resources for visual objects. J 
Neurosci. 2011; 31:8502–8511. [PubMed: 21653854] 
49. Shapiro KL, Miller CE. The role of biased competition in visual short term memory. 
Neuropsychologia. 2011; 49:1506–1517. [PubMed: 21335016] 
50. Bays MP, et al. Temporal dynamics of encoding, storage, and reallocation of visual working 
memory. J Vis. 2011; 11:6. [PubMed: 21911739] 
51. Curtis CE, D’Esposito M. Persistent activity in the prefrontal cortex during working memory. 
Trends Cogn Sci. 2003; 7:415–423. [PubMed: 12963473] 
52. Postle BR. Working memory as an emergent property of the mind and brain. Neuroscience. 2006; 
139:23–28. [PubMed: 16324795] 
53. Funahashi S. Space representation in the prefrontal cortex. Progr Neurobiol. (in press). 
54. Kastner S, et al. Topographic maps in human frontal cortex revealed in memory-guided saccade 
and spatial working-memory tasks. J Neurophysiol. 2007; 97:3494–3507. [PubMed: 17360822] 
55. Funahashi S, et al. Dorsolateral prefrontal lesions and oculomotor delayed-response performance: 
evidence for mnemonic ‘scotomas’. J Neurosci. 1993; 13:1479–1497. [PubMed: 8463830] 
56. Kanwisher N. Functional specificity in the human brain: a window into the functional architecture 
of the mind. Proc Natl Acad Sci U S A. 2010; 107:11163–11170. [PubMed: 20484679] 
57. Chang EF, et al. Categorical speech representation in human superior temporal gyrus. Nat 
Neurosci. 2010; 13:1428–1432. [PubMed: 20890293] 
58. Peelen MV, et al. Supramodal representations of perceived emotions in the human brain. J 
Neurosci. 2010; 30:10127–10134. [PubMed: 20668196] 
59. Caramazza A, Mahon BZ. The organization of conceptual knowledge: the evidence from category-
specific semantic deficits. Trends Cogn Sci. 2003; 7:354–361. [PubMed: 12907231] 
60. Anderson MC, et al. Remembering can cause forgetting: retrieval dynamics in long term memory. J 
Exp Psychol Learn. 1994; 20:1063–1087.
61. Carr TH, Dagenbach D. Semantic priming and repetition priming from masked words: Evidence 
for a center-surround attentional mechanism in perceptual recognition. J Exp Psychol. 1990; 
16:341–350.
62. Healey MK, et al. Direct evidence for the role of inhibition in resolving interference in memory. 
Psychol Sci. 2010; 21:1464–1470. [PubMed: 20807896] 
63. Schoner, G., et al. Dynamic Thinking: A Primer on Dynamic Field Theory. Oxford University 
Press; (in press)
64. Johnson JS, et al. Dynamic neural field model of visual working memory and change detection. 
Psychol Sci. 2009; 20:568–577. [PubMed: 19368698] 
65. Buonomano DV, Merzenich MM. Cortical plasticity: from synapses to maps. Annu Rev Neurosci. 
1998; 21:149–186. [PubMed: 9530495] 
66. Albright TD, et al. Columnar organization of directionally selective cells in visual area MT of the 
macaque. J Neurophysiol. 1984; 51:16–31. [PubMed: 6693933] 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 13
67. Da Costa S, et al. Human primary auditory cortex follows the shape of Heschl’s Gyrus. J Neurosci. 
2011; 31:14067–14075. [PubMed: 21976491] 
68. Atiani S, et al. Task difficulty and performance induce diverse adaptive patterns in gain and shape 
of primary auditory cortical receptive fields. Neuron. 2009; 61:467–480. [PubMed: 19217382] 
69. Tanaka K. Columns for complex visual object features in the inferotemporal cortex: Clustering of 
cells with similar but slightly different stimulus selectivities. Cereb Cortex. 2003; 13:90–99. 
[PubMed: 12466220] 
70. Wang G, et al. Optical imaging of functional organization in the monkey inferotemporal cortex. 
Science. 1996; 272:1665–1668. [PubMed: 8658144] 
71. Arzi A, Sobel N. Olfactory perception as a compass for olfactory neural maps. Trends Cogn Sci. 
2011; 15:537–545. [PubMed: 22001868] 
72. Auffarth B, et al. Continuous spatial representations in the olfactory bulb may reflect perceptual 
categories. Front Syst Neurosci. 2011; 5:82. [PubMed: 22013415] 
73. Knudsen EI, et al. Computational maps in the brain. Annu Rev Neurosci. 1987; 10:41–65. 
[PubMed: 3551761] 
74. Gould HJ III, et al. The relationship of corpus callosum connections to electrical stimulation maps 
of motor, supplementary motor, and the frontal eye fields in owl monkeys. J Comput Neurol. 1986; 
247:297–325.
75. Loach D, et al. An attentional mechanism for selecting appropriate actions afforded by graspable 
objects. Psychol Sci. 2008; 19:1253–1257. [PubMed: 19121133] 
76. Purves D, et al. Iterated patterns of brain circuitry (or how the cortex gets its spots). Trends 
Neurosci. 1992; 15:362–368. [PubMed: 1279855] 
77. Kaas JH. Topographic maps are fundamental to sensory processing. Brain Res Bull. 1997; 44:107–
112. [PubMed: 9292198] 
78. Knudsen EI. Auditory and visual maps of space in the optic tectum of the owl. J Neurosci. 1982; 
2:1177–1194. [PubMed: 7119872] 
79. Luck SJ, Vogel EK. The capacity of visual working memory for features and conjunctions. Nature. 
1997; 390:279–281. [PubMed: 9384378] 
80. Zhang W, Luck SJ. The number and quality of representations in working memory. Psychol Sci. 
2011; 22:1434–1441. [PubMed: 21987693] 
81. Wheeler M, Treisman AM. Binding in short-term visual memory. J Exp Psychol Gen. 2002; 
131:48–64. [PubMed: 11900102] 
82. Fougnie D, et al. What are the units of storage in visual working memory? J Vis. 2010; 10:1–11.
83. Bays PM, Husain M. Dynamic shifts of limited working memory resources in human vision. 
Science. 2008; 321:851. [PubMed: 18687968] 
84. Bays PM, et al. The precision of visual working memory is set by allocation of a shared resource. J 
Vis. 2009; 9:1–11. [PubMed: 19810788] 
85. Wilken P, Ma WJ. A detection theory account of change detection. J Vis. 2004; 4:1120–1135. 
[PubMed: 15669916] 
86. Xu Y, Chun MM. Dissociable neural mechanisms supporting visual short-term memory for objects. 
Nature. 2006; 440:91–95. [PubMed: 16382240] 
87. Brady TF, et al. A review of visual memory capacity: beyond individual items and towards 
structured representations. J Vis. 2011; 11:1–34.
88. Brady, TF.; Tenenbaum, JB. Encoding higher-order structure in visual working memory: A 
probabilistic model. In: Ohlsson, S.; Catrambone, R., editors. Proceedings of the 32nd Annual 
Conference of the Cognitive Science Society. Cognitive Science Society; 2010. p. 411-416.
89. O’Reilly RC, Rudy JW. Conjunctive representations in learning and memory: principles of cortical 
and hippocampal function. Psychol Rev. 2001; 108:311–345. [PubMed: 11381832] 
90. Treisman AM, Gelade G. A feature-integration theory of attention. Cogn Psychol. 1980; 12:97–
136. [PubMed: 7351125] 
91. Gray CM, Singer W. Stimulus-specific neuronal oscillations in orientation columns of cat visual 
cortex. Proc Natl Acad Sci U S A. 1989; 86:1698–1702. [PubMed: 2922407] 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 14
92. Bays PM, et al. Storage and binding of object features in visual working memory. 
Neuropsychologia. 2011; 49:1622–1631. [PubMed: 21172364] 
93. Fougnie D, Alvarez GA. Object features fail independently in visual working memory: evidence 
for a probabilistic feature-store model. J Vis. 2011; 11:1–12.
94. Randall, L. Ecco. 2005. Warped Passages: Unraveling the Mysteries of the Universe’s Hidden 
Dimensions. 
95. Wandell BA, et al. Visual field maps in human cortex. Neuron. 2007; 56:366–383. [PubMed: 
17964252] 
96. Alvarez GA. Representing multiple objects as an ensemble enhances visual cognition. Trends Cogn 
Sci. 2011; 15:122–131. [PubMed: 21292539] 
97. Farah M. Why does the somatosensory homunculus have hands next to face and feet next to 
genitals: a hypothesis. Neur Comput. 1998; 10:1983–1985.
98. Konkle T, Oliva A. A real-world size organization of object responses in occipitotemporal cortex. 
Neuron. 2012; 74:1114–1124. [PubMed: 22726840] 
99. Pylyshyn ZW, Storm. Tracking multiple independent targets: evidence for a parallel tracking 
mechanism. Spat Vis. 1988; 3:179–197. [PubMed: 3153671] 
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 15
Box 3
Competitive maps and the binding problem
A dinner party presents strong demands on cognitive resources. The limitations on 
encoding and remembering new faces and new names might be explained through 
competition within maps for various properties. However, what about matching those 
new faces, which are stored in one set of maps, to those new names, stored in another set 
of maps? This ‘binding problem’ challenges all models of perception, attention, and 
memory, and there are several classes of proposed mechanisms that might link separate 
representations together (e.g., [89–91]). It is likely that the strength of these links, as well 
as the competition for space that we are discussing here, both affect the limit in the 
number of items that can be attended, planned, or remembered. Nonetheless, there are 
aspects of the map architecture that are particularly relevant to the binding problem.
In some ways, maps are especially prone to binding difficulties, because their maximum 
of two dimensions limits the scope of what they can represent. Thus, strongly hierarchical 
representations must be linked across many maps, for instance, across the levels of the 
ventral object recognition hierarchy. Maintaining binding across such maps might be 
particularly demanding and recent evidence suggests that items are not actually stored as 
bound units in working memory [92,93].
However, in other ways, maps have properties that might help address these binding 
difficulties. First, note that some maps contain embedded dimensions (e.g., orientation or 
color for primary visual cortex; Box 1) arrayed in fine-scale local topologies. Embedded 
dimensions exist in early visual representations, such as V1, and are proposed in other 
disciplines (e.g., string theory; [94]). Second, in many cases, maps that share a common 
format allow simple cross-referencing. For example, target-related activity peaks in the 
retinotopic saccade and attention maps indicate the locations of the targets, but they do 
not carry information about their features or identity. Nevertheless, those locations point 
to features that can be found within other retinotopically organized visual areas [20,95]. 
Indeed, the classic feature integration theory binding model proposed location as the 
common property that links features across separate maps [90].
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 16
Figure 1. 
At top, a competitive map representation within a bounded, two-dimensional space. (a) This 
panel depicts how each location represents value in the space, with items represented by 
peaks of activation. Each peak also suppresses surrounding space, inhibiting nearby 
competitors [8]. This surround suppression will limit the number of items that can be 
simultaneously maintained. If the item spacing is dense, as in (b), the space will be 
inefficiently used. If the spacing is sparse, as in (c), the space is efficiently used and capacity 
is maximized, though still limited (by approximately six items in this case). (d) This panel 
shows that anatomical boundaries (e.g., the visual hemifield divisions of V1) can mimic 
spacing effects by eliminating mutual inhibition. At bottom, a slot representation limited to 
four independent items.
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 17
Figure 2. 
(a) Architecture of spatial attention (adapted from [19]). A network of areas form a 
competitive target map that subserves spatial attention, as well as eye movements. Peaks of 
activity specify retinotopic coordinates of feature data in earlier visual cortices which are 
shown, highly simplified, as a stack of aligned areas divided into right and left hemifields 
with the fovea in the center. In object recognition areas, cells have large receptive fields 
shown here as a heavy black outline for the receptive field of one cell that specializes in 
identifying corkscrews. These cells must rely on attention to bias input in favor of the target 
and suppress surrounding distractors, so that only a single item falls in the receptive field at 
any one time. The surround suppression must be imposed in early retinotopic areas, because 
the large fields in object recognition cannot locally modulate sensitivity. (b) Resource limits 
in multiple object tracking (MOT) tasks. In MOT, a participant is asked to track multiple 
moving objects (marked here in red for illustration only) among visually identical 
distractors, which requires constant spatial selection of those objects. When concurrent 
MOT displays are arranged within visual quadrants, tracking within two vertically arranged 
displays leads to ‘resource drains’, where performance drops. However, when arranged 
horizontally, resources are ‘independent’, because performance is virtually unaffected [33]. 
At bottom, the flexible map account predicts this effect, because the visual hemifield 
boundary strongly blocks inhibition horizontally, but only weakly blocks it vertically [28]. A 
strong competitive map account of such effects predicts that almost all performance 
limitations in this task can be ascribed to competition within a spatial map representing 
target positions [10].
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Franconeri et al. Page 18
Figure 3. 
Maps as limits on visual memory capacity. At left, data from [38], showing that VSTM can 
hold fewer objects as they become more complex, with an ultimate limit of approximately 
four total objects that can be held. The remaining boxes present a series of maps that might 
create visual short-term memory limitations. The first set shows two types of visual displays, 
complex shapes and simple letters. The next column depicts spatial selection maps (though 
feature selection, e.g., by color, is equally likely). In this case, one or two locations are 
selected, biasing competition within multiple hierarchical levels of visual data maps (V1–
V4, MT, IT) relevant to recognition of those objects. Critically, because the space for 
complex shapes is more densely packed and/or requires simultaneous activation of more 
locations to encode the complex shape information, few shapes can be represented 
concurrently without their representations degrading, and therefore only one shape should be 
attended at once in the selection maps. Letters, by contrast, have a well-spaced map of high-
level identities created by vast experience, and, therefore, multiple letter identities can be 
reliably encoded at once, allowing multiple locations to be attended at once in the selection 
maps. On the right, a hypothetical ‘spatial’ memory map that holds pointers to previously 
seen visual data (or more likely, pointers to selection maps that point to those data). These 
could be subserved by connections between activation maintenance structures in frontal 
cortex and parietal selection maps.
Trends Cogn Sci. Author manuscript; available in PMC 2016 October 03.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript