In search of common foundations for cortical computation

It is worthwhile to search for forms of coding, processing, and learning common to various cortical regions and cognitive functions. Local cortical processors may coordinate their activity by maximizing the transmission of information coherently related to the context in which it occurs, thus forming synchronized population codes. This coordination involves contextual field (CF) connections that link processors within and between cortical regions. The effects of CF connections are distinguished from those mediating receptive field (RF) input; it is shown how CFs can guide both learning and processing without becoming confused with the transmission of RF information. Simulations explore the capabilities of networks built from local processors with both RF and CF connections. Physiological evidence for synchronization, CFs, and plasticity of the RF and CF connections is described. Coordination via CFs is related to perceptual grouping, the effects of context on contrast sensitivity, amblyopia, implicit influences of color in achromotopsia, object and word perception, and the discovery of distal environmental variables and their interactions through self-organization. Cortical computation could thus involve the flexible evaluation of relations between input signals by locally specialized but adaptive processors whose activity is dynamically associated and coordinated within and between regions through specialized contextual connections.     

The colour of Os: naturally biased associations between shape and colour

Many letters of the alphabet are consistently mapped to specific colours by English-speaking adults, both in the general population and in individuals with grapheme-colour synaesthesia who perceive letters in colour. Such associations may be naturally biased by intrinsic sensory cortical organisation, or may be based in literacy (eg 'A' is for 'apple', apples are red; therefore A is red). To distinguish these two hypotheses, we tested pre-literate children in three experiments and compared their results to those of literate children (aged 7-9 years) and adults. The results indicate that some colour letter mappings (O white, X black) are naturally biased by the shape of the letter, whereas others (A red, G green) may be based in literacy. They suggest that sensory cortical organisation initially binds colour to some shapes, and that learning to read can induce additional associations, likely through the influence of higher-order networks as letters take on meaning.     

Cerebral achromatopsia: colour blindness despite wavelength processing

Cortical colour blindness is caused by brain damage to the ventro-medial occipital and temporal lobes. A possible explanation is that the pathway responsible for transmitting information about wavelength and its subsequent elaboration as colour has been destroyed at the cortical level. However, several signs of chromatic processing persist in an achromatopsic subject who, despite his inability to tell colours apart, can still detect chromatic borders, perceive shape from colour, and discriminate the direction in which a striped pattern moves when the determination of direction requires the viewer to 'know' which stripes have a particular colour. Perhaps only the information about wavelength that leads to conscious awareness of colour has been destroyed. It is unclear whether incomplete achromatopsia is merely a less severe form of the disorder or whether it is qualitatively different, perhaps reflecting impaired colour constancy. In monkeys, removing cortical area V4 impairs performance on colour constancy tasks but, invariably, impairs several other aspects of visual perception. If the lesion that causes total achromatopsia in human subjects corresponds to area V4 in monkeys, it is an unsolved puzzle that a totally achromatopsic subject paradoxically demonstrates certain characteristics of colour constancy, unless his residual performance reflects the much underrated retinal contribution to colour constancy.     

Natural scenes can be identified as rapidly as individual features

Can observers determine the gist of a natural scene in a purely feedforward manner, or does this process require deliberation and feedback? Observers can recognise images that are presented for very brief periods of time before being masked. It is unclear whether this recognition process occurs in a purely feedforward manner or whether feedback from higher cortical areas to lower cortical areas is necessary. The current study revealed that the minimum presentation time required to identify or to determine the gist of a natural scene was no different from that required to determine the orientation or colour of an isolated line. Conversely, a visual task that would be expected to necessitate feedback (determining whether an image contained exactly six lines) required a significantly greater minimum presentation time. Assuming that the orientation or colour of an isolated line can be determined in a purely feedforward manner, these results indicate that the identification and the determination of the gist of a natural scene can also be performed in a purely feedforward manner. These results challenge a number of theories of visual recognition that require feedback.     

SLAM: a connectionist model for attention in visual selection tasks

SLAM, the SeLective Attention Model, performs visual selective attention tasks, an analysis of which shows that two processes, object and attribute selection, are both necessary and sufficient. It is based upon the McClelland and Rumelhart (1981) model for visual word recognition, with the addition of a response selection and evaluation mechanism. The responses may be correct or incorrect and, in particular conditions, SLAM may not make a response at all. Moreover, it allows for the generation of specific responses in time. SLAM's main characteristics are parallelism restricted by competition within modules, heterarchical processing in a hierarchical structure, and generation of responses as a result of relaxation given the conjoint constraints of stimulation, object, and attribute selection. The model is considered to represent an individual subject performing filtering tasks and demonstrates appropriate selective behavior. It is also tested quantitatively using a single tentative set of model parameters. The study reports simulations of four different filtering experiments, modeling response latencies, and error proportions. Specifications are made to take account of instructions, previous trials, and the effect of a barmarker cue and of asynchronies in stimulus and cue onsets. The model is then extended in order to provide simulations of a number of Stroop experiments, which can be regarded as filtering tasks with nonequivalent stimuli. The extension required for Stroop simulations is the addition of direct connections between compatible stimulus and response aspects. The direct connections do not affect the simulation of simpler filtering tasks. A variety of different experiments carried out by different authors is simulated. The model is discussed in terms of how modular architecture and the interaction of excitation and inhibition generate facilitation or inhibition of response latencies.     

Rigorous results,cross-model justification,and the transfer of empirical warrant: the case of many-body models in physics

This paper argues that a successful philosophical analysis of models and simulations must accommodate an account of mathematically rigorous results. Such rigorous results may be thought of as genuinely model-specific contributions, which can neither be deduced from fundamental theory nor inferred from empirical data. Rigorous results provide new indirect ways of assessing the success of models and simulations and are crucial to understanding the connections between different models. This is most obvious in cases where rigorous results map different models on to one another. Not only does this put constraints on the extent to which performance in specific empirical contexts may be regarded as the main touchstone of success in scientific modelling, it also allows for the transfer of warrant across different models. Mathematically rigorous results can thus come to be seen as not only strengthening the cohesion between scientific strategies of modelling and simulation, but also as offering new ways of indirect confirmation.     

Generative models, brain function and neuroimaging

The representational capacity and inherent function of any neuron, neuronal population or cortical area in the brain is dynamic and context-sensitive. Functional integration, or interactions among brain systems, that employ driving (bottom up) and backward (top-down) connections, mediate this adaptive and contextual specialisation. A critical consequence is that neuronal responses, in any given cortical area, can represent different things at different times. This can have fundamental implications for the design of brain imaging experiments and the interpretation of their results. Our arguments are developed under generative models of brain function, where higher-level systems provide a prediction of the inputs to lower-level regions. Conflict between the two is resolved by changes in the higher-level representations, which are driven by the ensuing error in lower regions, until the mismatch is "cancelled". From this perspective the specialisation of any region is determined both by bottom-up driving inputs and by top-down predictions. Specialisation is therefore not an intrinsic property of any region but depends on both forward and backward connections with other areas. Because the latter have access to the context in which the inputs are generated they are in a position to modulate the selectivity or specialisation of lower areas. The implications for classical models (e.g., classical receptive fields in electrophysiology, classical specialisation in neuroimaging and connectionism in cognitive models) are severe and suggest these models may provide incomplete accounts of real brain architectures. Here we focus on the implications for cognitive neuroscience in the context of neuroimaging.     

Interhemispheric effects of simulated lesions in a neural model of single-word reading

A neural model consisting of paired cerebral hemispheric regions interacting via homotopic callosal connections was trained to generate pronunciations for 50 monosyllabic words. Lateralization of this task occurred readily when different underlying cortical asymmetries were present. Following simulated focal cortical lesions of systematically varied sizes, acute changes in the distribution of cortical activation were found to be most consistent with experimental data when interhemispheric interactions were assumed to be excitatory. During subsequent recovery, the contribution of the unlesioned hemispheric region to performance improvement was a function of both the amount of preexisting lateralization and the side and size of the lesion. These results are discussed in the context of unresolved issues concerning the mechanisms underlying language lateralization, the nature of interhemispheric interactions, and the role of the nondominant hemisphere in recovery from adult aphasia.     

The influence of depth segmentation on colour constancy

In real scenes, surfaces in different depth planes often differ in the luminance and chromatic content of their illumination. Scene segmentation is therefore an important issue when considering the compensation of illumination changes in our visual perception (lightness and colour constancy). Chromatic adaptation is an important sensory component of colour constancy and has been shown to be linked to the two-dimensional spatial structure of a scene (Werner, 2003 Vision Research 43 1611 - 1623). Here, the question is posed whether this cooperation also extends to the organisation of a scene in depth. The influence of depth on colour constancy was tested by introducing stereo disparity, whereby the test patch and background were perceived in either the same or one of five different depth planes (1.9-57 min of arc). There were no additional cues to depth such as shadows or specular highlights. For consistent illumination changes, colour constancy was reduced when the test patch and background were separated in depth, indicating a reduction of contextual influences. An interaction was found between the influences of stereo depth and spatial frequency on colour constancy. In the case of an inconsistent illumination change, colour constancy was reduced if the test patch and background were in the same depth plane (2-D condition), but not if they were separated in depth (3-D condition). Furthermore, colour constancy was slightly better in the 3-D inconsistent condition than in the 2-D inconsistent condition. It is concluded that depth segmentation supports colour constancy in scenes with inconsistent illumination changes. Processes of depth segmentation are implemented at an early sensory stage of colour constancy, and they define visual regions within which the effects of illuminant changes are discounted for separately. The results support recent models that posit such implementation of scene segmentation in colour constancy.     

Evidence of enhanced self-organization in the work-hardening stage V of fcc metals

Stages IV and V of the stress-strain curves of fcc metals, τ = τ (γ), are shownto be clearly discernible in the measurements of Taylor and Quinney onpolycrystalline copper strongly deformed at room temperature. The differentialstored-energy fraction determined by these workers may serve as an indicator ofthe degree of self-organization during the deformation process and hence of thedislocation mechanisms involved. In the data there is no indication of a change inmechanism at the stress at which the slope of the work-hardening curve becomesstrain independent, conventionally associated with the onset of stage IV. Only athigher stresses is there evidence of enhanced self-organization and hence of amechanism that diOEers significantly from stage-III hardening. It is argued thatthe enhanced self-organization is related to the onset of stage V, characterized byvanishing work-hardening. In this context, a basic difficulty in obtainingquantitative information on stage V from the widely used torsion tests ispointed out.     

Perception of chromatic motion requires luminance interaction

There is an ongoing debate related to whether chromatic motion perception arises as a consequence of a chromatic signal only (eg Wandell et al 1999 Neuron 24 901-909) or a signal that is essentially based on luminance processes (luminance artifacts) (Mullen et al 2003 Vision Research 43 1235-1247). These two views conform to the idea that colour and luminance processes are physiologically independent (Livingstone and Hubel 1988 Science 240 740-749), but according to other reports many primary cortical 'V1' cells respond to both colour and luminance contrast (eg Vidyasagar et al 2002 European Journal of Neuroscience 16 945-956). A psychophysical task was designed to test whether possible interaction between luminance and chromatic contrast could account for perception of chromatic motion. It is shown that subjects respond in a manner that reflects involvement of both processes.     

Sources of somatosensory input to the caudal belt areas of auditory cortex

The auditory cortex of nonhuman primates is comprised of a constellation of at least twelve interconnected areas distributed across three major regions on the superior temporal gyrus: core, belt, and parabelt. Individual areas are distinguished on the basis of unique profiles comprising architectonic features, thalamic and cortical connections, and neuron response properties. Recent demonstrations of convergent auditory-somatosensory interactions in the caudomedial (CM) and caudolateral (CL) belt areas prompted us to pursue anatomical studies to identify the source(s) of somatic input to auditory cortex. Corticocortical and thalamocortical connections were revealed by injecting neuroanatomical tracers into CM, CL, and adjoining fields of marmoset (Callithrix jacchus jacchus) and macaque (Macaca mulatta) monkeys. In addition to auditory cortex, the cortical connections of CM and CL included somatosensory (retroinsular, Ri; granular insula, Ig) and multisensory areas (temporal parietal occipital, temporal parietal temporal). Thalamic inputs included the medial geniculate complex and several multisensory nuclei (suprageniculate, posterior, limitans, medial pulvinar), but not the ventroposterior complex. Injections of the core (A1, R) and rostromedial areas of auditory cortex revealed sparse multisensory connections. The results suggest that areas Ri and Ig are the principle sources of somatosensory input to the caudal belt, while multisensory regions of cortex and thalamus may also contribute. The present data add to growing evidence of multisensory convergence in cortical areas previously considered to be 'unimodal', and also indicate that auditory cortical areas differ in this respect.     

Understanding the nature of the general factor of intelligence: the role of individual differences in neural plasticity as an explanatory mechanism

The nature of the general factor of intelligence, or g, is examined. This article begins by observing that the finding of a general factor of intelligence appears to be inconsistent with current findings in neuroscience and cognitive science, where specific connections are argued to be critical for different intellectual abilities and the brain is argued to develop these connections in response to environmental stimuli. However, it is then observed that if people differed in neural plasticity, or the ability to adapt their connections to the environment, then those highly developed in one intellectual ability would be highly developed in other intellectual abilities as well. Simulations are then used to confirm that such a pattern would be obtained. Such a model is also shown to account for many other findings in the field of intelligence that are currently unexplained. A critical period for intellectual development is then emphasized.     

Transfer of visual information after unilateral input to the brain

Visual callosal connections are more numerous and widespread in the association areas than in the primary visual cortex and adjoining visual areas. In keeping with this, the amount of "wrong" ipsilateral visual field that is represented in the various cortical areas of primates and cats increases as one goes from primary visual cortex to extraoccipital areas. Therefore it can be argued that transfer of unilaterally presented visual stimuli occurs mainly at the temporal and parietal cortical level.     

Enhanced cholinergic suppression of previously strengthened synapses enables the formation of self-organized representations in olfactory cortex

Computational modeling assists in analyzing the specific functional role of the cellular effects of acetylcholine within cortical structures. In particular, acetylcholine may regulate the dynamics of encoding and retrieval of information by regulating the magnitude of synaptic transmission at excitatory recurrent connections. Many abstract models of associative memory function ignore the influence of changes in synaptic strength during the storage process and apply the effect of these changes only during a so-called recall-phase. Efforts to ensure stable activity with more realistic, continuous updating of the synaptic strength during the storage process have shown that the memory capacity of a realistic cortical network can be greatly enhanced if cholinergic modulation blocks transmission at synaptic connections of the association fibers during the learning process. We here present experimental data from an olfactory cortex brain slice preparation showing that previously potentiated fibers show significantly greater suppression (presynaptic inhibition) by the cholinergic agonist carbachol than unpotentiated fibers. We conclude that low suppression of non-potentiated fibers during the learning process ensures the formation of self-organized representations in the neural network while the higher suppression of previously potentiated fibers minimizes interference between overlapping patterns. We show in a computational model of olfactory cortex, that, together, these two phenomena reduce the overlap between patterns that are stored within the same neural network structure. These results further demonstrate the contribution of acetylcholine to mechanisms of cortical plasticity. The results are consistent with the extensive evidence supporting a role for acetylcholine in encoding of new memories and enhancement of response to salient sensory stimuli.     

Valuable orientations capture attention

Visual attention has long been known to be drawn to stimuli that are physically salient or congruent with task-specific goals. Several recent studies have shown that attention is also captured by stimuli that are neither salient nor task relevant, but that are rendered in a colour that has previously been associated with reward. We investigated whether another feature dimension—orientation—can be associated with reward via learning and thereby elicit value-driven attentional capture. In a training phase, participants received a monetary reward for identifying the colour of Gabor patches exhibiting one of two target orientations. A subsequent test phase in which no reward was delivered required participants to search for Gabor patches exhibiting one of two spatial frequencies (orientation was now irrelevant to the task). Previously rewarded orientations robustly captured attention. We conclude that reward learning can imbue features other than colour—in this case, specific orientations—with persistent value.     

Context-sensitive binding by the laminar circuits of V1 and V2: A unified model of perceptual grouping,attention, and orientation contrast

A detailed neural model is presented of how the laminar circuits of visual cortical areas V1 and V2 implement context-sensitive binding processes such as perceptual grouping and attention. The model proposes how specific laminar circuits allow the responses of visual cortical neurons to be determined not only by the stimuli within their classical receptive fields, but also to be strongly influenced by stimuli in the extra-classical surround. This context-sensitive visual processing can greatly enhance the analysis of visual scenes, especially those containing targets that are low contrast, partially occluded, or crowded by distractors. We show how interactions of feedforward, feedback, and horizontal circuitry can implement several types of contextual processing simultaneously, using shared laminar circuits. In particular, we present computer simulations that suggest how top-down attention and preattentive perceptual grouping, two processes that are fundamental for visual binding, can interact, with attentional enhancement selectively propagating along groupings of both real and illusory contours, thereby showing how attention can selectively enhance object representations. These simulations also illustrate how attention may have a stronger facilitatory effect on low contrast than on high contrast stimuli, and how pop-out from orientation contrast may occur. The specific functional roles which the model proposes for the cortical layers allow several testable neurophysiological predictions to be made. The results presented here simulate only the boundary grouping system of adult cortical architecture. However, we also discuss how this model contributes to a larger neural theory of vision that suggests how intracortical and intercortical feedback help to stabilize development and learning within these cortical circuits. Although feedback plays a key role, fast feedforward processing is possible in response to unambiguous information. Model circuits are capable of synchronizing quickly, but context-sensitive persistence of previous events can influence how synchrony develops. Although these results focus on how the interblob cortical processing stream controls boundary grouping and attention, related modelling of the blob cortical processing stream suggests how visible surfaces are formed, and modelling of the motion stream suggests how transient responses to scenic changes can control long-range apparent motion and also attract spatial attention.     

Boundary conditions on parallel processing in human vision

A new theory of visual search is tested experimentally with simple colour patches. The essential element of this new theory is that, whatever the search materials, efficiency increases continuously with (i) decreasing similarity between targets and nontargets, and (ii) increasing similarity between one nontarget and another. Control of 'attention' (access to visual short-term memory) is seen as a competitive interaction between display elements, and the theory shows how stimulus similarities influence the outcome of this competition. One alternative view is that parallel visual processes are limited to local mismatch detection. Search is parallel if the target forms a break in an otherwise homogeneous field, but is serial when absolute stimulus identification is required. It is shown, however, that even colour identification can be parallel, providing targets and nontargets are sufficiently dissimilar. A second alternative view is that search for simple features is parallel whereas search for conjunctions is serial. Conjunction search, however, has a characteristic similarity structure: different kinds of nontarget each share one relevant attribute with the target, but none with one another. When this structure is mimicked in search for colour patches, correspondingly poor performance is obtained.