Speaker Abstracts

Visual & Cognitive Circuits (Liz Romanski, Session Chair)

Simon Thorpe, CNRS, France

From vision to action in 100 ms – is there time for cognition?

When two natural scenes are displayed left and right of fixation, human subjects can initiate saccades towards the side containing an animal target in just 120-30 ms (Kirchner & Thorpe, 2006, Vision Research, 46, 1762), but processing time is even faster when the target is a human face. Accuracy can be around 90% even when the mean saccadic reaction time is only 140 ms, and the fastest saccades can start 100-110 ms after image onset. If we allow 20 ms for response initiation, it would appear that the underlying visual processing can be done in as little as 80 ms. What sorts of visual representations could be involved? How might the brain produce anything that is selective for faces at such short latencies? Recent modeling work using feed-forward networks coupled with Spike-Time Dependent Plasticity has shown how unsupervised learning schemes can lead to the development of selectivity to frequently encounter visual patterns (Masquelier & Thorpe, 2007, PLoS Comput Biol, 3, e31). Could the fact that faces often dominate the visual environment of newborn babies allow neurons in relatively early visual areas such as V4 to become selective for facial features? If so, this might help explain how such rapid behavioral responses can be generated, but it leaves open the question of whether this involves cognition.

Close Window

Emilio Salinas, Wake Forest University

How behavioral constraints influence sensory tuning curves

Many types of neurons respond to sensory stimuli such that their response profiles have a single peak at a 'preferred' point, and population codes are often thought of as arrays of such units. However, many neurons show monotonic dependencies on sensory parameters. Such monotonic responses are both widespread and difficult to reconcile with current ideas about sensory coding, which are based on two fundamental ingredients: stimulus statistics and efficiency criteria (redundancy reduction, decorrelation, sparseness, etc.). I will discuss a complementary point of view in which the responses of a neuronal population are evaluated in terms of the range of outputs that they are capable of generating. The idea is that optimal sensory representations should take into account not only the statistics of the sensory world, but also the statistics of the downstream motor activity that generates behavior. When the downstream responses are non-monotonic, in general the sensory responses that are optimal for driving them will have peaks. However, if the downstream responses are monotonic, then the optimal sensory tuning curves will be predominantly monotonic. Biological examples that are consistent with these results include the encoding of binocular disparity and of heading direction in the visual cortex, as well as the encoding of echo delay in bats. These results suggest that knowledge about the downstream impact of sensory representations is crucial for understanding some of their basic properties.

Close Window

Helen Barbas, Boston University

How the frontal lobe controls the mind's eye

Successful completion of everyday tasks requires focusing on relevant information and ignoring irrelevant stimuli. The prefrontal cortex in primates has a key role in these processes through mechanisms that are not well understood. Recent findings indicate that the prefrontal cortex may select or ignore stimuli through pathways that form synapses with laminar-specific excitatory and inhibitory neurons in sensory and high-order association cortices. In addition, prefrontal pathways target extensively the frontal and sensory sectors of the inhibitory thalamic reticular nucleus, and the inhibitory intercalated masses of the amygdala. Circuit-based models suggest that these prefrontal pathways may select relevant sensory stimuli and suppress distracters at an early stage of processing. The interface of the prefrontal cortex with subcortical and cortical inhibitory and excitatory systems provides the structural basis for suppressing irrelevant stimuli, allowing behaviorally significant signals to gain access to the cortex.

Close Window

Leslie Ungerleider, NIMH

The primate prefrontal cortex and the executive control of attention

The prefrontal cortex is thought to play a prominent role in the cognitive control of sensory information. To examine its contribution to the allocation of visual attention, we removed the prefrontal cortex unilaterally in combination with transection of the forebrain commisures in two adult rhesus monkeys. As a result, visual processing in only one hemisphere could be modulated by feedback inputs from the prefrontal cortex. Monkeys were trained to fixate a central spot and their performance in the affected and unaffected visual hemifields was compared on a variety of attention-demanding tasks. The results showed that the prefrontal cortex is critically involved in the top-down selection of the behaviorally relevant target from surrounding irrelevant distracters for processing resources and in updating this information from moment to moment. The prefrontal cortex does not appear to play a crucial role, however, when target selection from surrounding distracters is based on bottom-up mechanisms. Parallel neuroimaging studies in humans reveal the focus within human prefrontal cortex for the top-down attentional effects.

Close Window

Visual Signals in Cognitive Circuits (Greg DeAngelis, Session Chair)

Tatiana Pasternak, University of Rochester

Representation of visual motion during motion discrimination tasks in primate prefrontal cortex

I will show that during motion discrimination tasks, prefrontal cortex (PFC) neurons carry faithful representation of visual motion stimuli being discriminated. During a task requiring the monkeys to discriminate and remember stimulus direction, many neurons show direction selectivity and represent motion coherence in a way that is reminiscent of neurons in area MT. During a task requiring the discrimination of stimulus speed and ignoring its direction, the same neurons show tuning for stimulus speed resembling speed selectivity in MT. The nature and the temporal dynamics of these motion selective responses support their bottom-up origins. This response selectivity strongly depends on task demands. Thus, during speed discrimination, direction selectivity is reduced and delayed but not completely eliminated, suggesting a link between speed and direction signals expected from signals arriving from motion processing cortical neurons. In some neurons, the reduction in direction selectivity results from a decrease in the response to the preferred direction and in other neurons from an increase in the response to the anti-preferred direction, suggesting that PFC neurons possess or have access to the relatively low-level motion mechanisms. Finally, when the animals are not required to use motion stimuli to get a reward, responses to these stimuli become weak and stimulus selectivity nearly disappears. These observations demonstrate that PFC neurons, in addition to carrying signals about the rules governing the use of sensory stimuli, carry bottom-up signals that could allow active participation in sensory processing of behaviorally relevant stimuli.

Close Window

Andreas Nieder, University of Tuebingen, Germany

Representation of number in the primate brain

The verbal number concept allows humans to develop the superior mathematical skills that are a hallmark of technologically advanced cultures. Recent findings in animal cognition, developmental psychology and anthropology, however, indicate that numerical skills are rooted in non-verbal biological primitives. We investigated the neural foundations of quantitative categories and concepts in behaving macaque monkeys in combined psychophysical/neurophysiological studies. Monkeys were trained to discriminate different types of quantitative information in delayed match-to-sample tasks. Many neurons in the prefrontal and posterior parietal cortices encoded stimulus magnitude during sample presentation, or maintained this quantitative information ‘on line’ during a memory period. The tuning characteristics of such neurons can explain basic psychophysical effects in dealing with quantities (such as distance and size effects). Tuning to the preferred quantity was deteriorated whenever the monkeys made judgment errors, indicating the behavioral relevance of quantity-selective neurons. The current data shed light on the question of how the primate brain processes quantity information at an evolutionary early stage.

Close Window

John Assad, Harvard Medical School

Encoding behavioral relevance in parietal cortex

Flexible control of behavior requires the selective processing of task-relevant sensory information. A great deal of evidence suggests a central role for the parietal cortex in these functions. I will discuss the role of parietal cortex in the selective representation of visual information in behaviors for stimulus categorization, selective spatial attention and movement initiation. I will also show evidence for a  hierarchy in processing from lower to higher parietal cortical areas, with lower parietal areas providing a more veridical representation of the retinal stimulus (particularly with respect to motion), and higher parietal areas showing a more plastic representation that can be adapted according to the demands of the task at hand.

Close Window

Wendy Suzuki, New York University

Associative learning signals in the monkey medial temporal lobe

A critical function of the medial temporal lobe is the ability to successfully acquire new declarative information about facts and events that includes new associations between initially unrelated items (associative learning). A major goal of my lab is to understand the brain basis of new associative learning. I will first summarize the studies we have done characterizing the patterns of neural activity seen as monkeys learn new associations on-line. These studies have shown that neurons throughout the medial temporal lobe signal new learning with changes in their stimulus-selective response properties. More recent studies have revealed that these changes in stimulus-selective responses reflect the animal’s behavioral learning strategy. A surprising new finding shows that hippocampal neurons also provide a powerful signal of trial outcome, differentiating between trials that are correct or wrong. I will discuss the possible of role of these signals in a feedback process by which information about behavioral outcome can be used to strengthen correct performance and modify error performance.

Close Window

Cognitive Influences on Visual Processing I (Charles Duffy, Session Chair)

Marisa Carrasco, New York University

Spatial and feature-based attention: psychophysics and neuroimaging studies

We select relevant information by attending to particular locations and features in the environment. First, I will present psychophysical studies comparing spatial and feature-based attention. In one study, we address the temporal dynamics of spatial and feature-based attention with identical stimuli and task. We show that spatial attention is deployed faster than feature-based attention, but that at longer delays they both exert the same effect on the detection of a speed increment task. In another study, we compare the effects of spatial and feature-based attention on motion selective channels, using equivalent noise functions and deriving population responses. The finding that spatial attention reduces the threshold for direction of motion only at low levels of external noise (high motion coherence) suggests that this type of attention is mediated by a gain change in the population response. The finding that feature-based attention reduces the threshold across all levels of external noise (encompassing high and low motion coherence) suggests that it is mediated by both gain and tuning changes in the population response. Lastly, I will discuss two studies that relate psychophysical performance and fMRI activity. One study deals with the effects of covert spatial attention and contrast sensitivity. The other deals with feature-based attention and its effects on selective adaptation and the tilt aftereffect. Together these studies advance our understanding of the behavioral consequences and the neural correlates of spatial and feature-based attention.

Close Window

Frank Tong, Vanderbilt University

Decoding the contents of perception and attention from human brain activity

Is it possible to determine what a person is seeing, experiencing, or paying attention to, using noninvasive measures of brain activity? My lab has developed a novel decoding approach to extract information about orientation and motion direction from fMRI activity patterns obtained from the human visual cortex. Random variability in the distribution of feature-selective columns or neurons can lead to weak biases in individual fMRI voxels; by pooling the information available from many voxels we can obtain robust ensemble feature selectivity. Our studies indicate that we can reliably decode which of several orientations or motion directions a person is seeing from activity patterns in early visual areas. These feature-selective responses are strongly stimulus-driven. They depend on the strength and quality of the incoming visual signal, and can be observed even when the subjects must attend to irrelevant letter stimuli rapidly presented at fixation. Nonetheless, these orientation and direction-selective responses are also strongly biased by feature-based attention. We can reliably predict which of two overlapping orientations or motion directions a subject is paying attention to based on activity in early visual areas. Bias effects can even be found in V1, indicating that attention modulates feature-selective responses at the earliest possible stage of processing. Finally, when spatial attention is directed to one of two lateralized gratings, we find stronger feature-selective responses at the attended location. Our studies of population-level activity indicate that top-down effects of spatial and feature-based attention can dynamically alter the gain of feature-selective responses in early visual areas.

Close Window

David Heeger, New York University

The normalization model of attention

Attention has been reported to have a wide variety of effects on the responses of neurons on visual cortex. There is evidence that attention increases contrast sensitivity (described as a change in contrast gain), that it scales neuronal responses by a fixed gain factor (response gain), that it causes intermediate effects that appear consistent with neither contrast gain nor response gain, that it sharpens neuronal turning curves, and that it can, with multiple stimuli in the receptive field, reduce neuronal responses. These different effects of attentional modulation have not previously been explained within the framework of a single computational model. We describe a model of attention that exhibits each of these different forms of attentional modulation, depending on the stimulus conditions and the spread (or selectivity) of the attention field in the model. The model consists of three basic components: 1) the stimulus-evoked excitatory field, 2) divisive suppression that is pooled over a larger region of spatial locations and features (orientations) than the excitatory field, and 3) an attention field which is multiplied by the excitatory field (and inherited by the suppressive field), the gain of which is specified in terms of its spatial and featural extents. In addition to unifying a range of experimental data within a common computational framework, the proposed model helps reconcile proposals that have been taken to represent alternative models of attention. We argue that the ostensible variety and complexity of the results reported in the literature emerge from the variety of empirical protocols that were used, such that the results observed in any one experiment depended on the stimulus conditions and the attentional strategy, a notion that we define precisely in terms of the attention field in the model, but that has not typically been completely under experimental control.

Close Window

Stefan Treue, University of Göttingen, Germany

Attentional influences in the dorsal pathway: of space, features and objects

Close Window

Cognitive Influences on Visual Processing II (Duje Tadin, Session Chair)

Patrick Cavanagh, University of Paris, France

Attention and nonretinotopic processing in vision

Attention allows a desired target to be selected from a cluttered field of distractors. We examine the limits of selection in both space and time. In space, we estimate the central selection area of attention from measures of "crowding" whereas we use attentive tracking tasks to examine the suppressive area surrounding selection. We attribute capacity limits of visual attention to this mutual interference between suppressive surrounds of targets. We find the interference to be hemifield and quadrant limited suggesting an early quadrantic locus for the inter-target interference. Using moving attention, we study the integration of information across space when the target object moves but they eyes do not and find this object-based integration to have similar properties to transsaccadic integration when the eyes move and the object does not. Specifically, surface features like color and motion can be integrated across long distances but shape features like letter identity cannot. This result suggests that non-retinotopic integration (whether for moving objects or moving eyes) relies on crude summation by cells with large receptive fields, a process that cannot align information in object-centered coordinates and so only works for information that needs no alignment. The summation depends on attention to limit summation to properties of only the target object.

Close Window

Charles Gilbert, The Rockefeller University

Brain states

Vision is active. It is a dynamic process, resulting from an interaction between context, perceptual learning and top-down influences. All cortical and thalamic levels of sensory processing are subject to powerful top-down influences, the shaping of lower level processes by higher order, more complex information and cognitive states. There is an expanding view of the kind of information that is conveyed in a top-down fashion, including attention, expectation, and perceptual task. As a consequence every cortical area acts as an adaptive processor, undergoing continuing cycles of state change and functional switching, with earlier states influencing the way in which the bottom up sensory information is interpreted in subsequent states. Recording from primary visual cortex (V1) while animals perform shape detection and discrimination tasks, we find that neurons adapt different stimulus selectivities as animals learn these tasks. Moreover, neurons alter their functional properties along with the task requirements. The output from V1 therefore reflects both sensory and behavioral context, which may reflect an interaction between feedback pathways to V1 and local circuits within V1. We propose that during perceptual learning, both the encoding and recall of learned information involves a selection of the appropriate inputs that convey information about the stimulus being discriminated.

Close Window

Carl Olson, Carnegie Mellon University

Selectivity for features, conjunctions and configurations in ventral stream cortex

Neurons in monkey inferotemporal cortex (IT), when tested with digitized images from any arbitrary library, seem like snowflakes, all alike at one scale (in that they exhibit pattern selectivity), each unique at another scale (in that each responds to a distinctive subset of images). What accounts for this remarkable degree of pattern selectivity? By what properties are images that elicit a strong response from a given neuron set apart from other images that do not? I will describe experiments testing three hypotheses: that IT neurons are selective for

1. specific features
2. specific conjunctions of features
3. specific global arrangements of features

The results cast light on the computational underpinnings of pattern selectivity in IT. In so doing, they help to explain human performance in tasks requiring visual search and the perceptual discrimination of hierarchical figures.

Close Window

Shinsuke Shimojo, California Institute of Technology

Behavioral and neural correlates of visual preference decision making

Even a newborn Infant preferential looks at an object over others, which may be interpreted either as just an orienting response or a preference decision making (, or both). It also raises an intriguing question as to how orienting behavior and cognitive preference decision are related in adults. We found that the observer's gaze is biased towards the to-be-chosen face as more attractive, long before (s)he is consciously aware of the decision ("gaze cascade effect"). Moreover, (a) we could manipulate the observer's preference decision by manipulating gaze (and it could not be attributed to the mere exposure effect), (b) when the observer was allowed only local visual sampling through a small window which moved along with gaze direction, the gaze bias started surprisingly early, during the local sampling, and (c) the qualitatively similar gaze cascade effect was found in stimuli other than faces (geometric figures, jewelry, watches, etc.). These all strongly argue that orienting response is involved critically in the process of preference decision making.

Our fMRI study employing a similar face preference task indicated the nucleus accumbens, the orbito-frontal cortex, and the operculum/insula activation in this dynamic time sequence. Moreover, the initial activity in the nucleus accumbens were independent of the explicit task, reflecting task-irelevant "first impression." Yet another issue about preference is how memory modulates preference. The literature was split into the novelty principle vs. the familiarity principle, with very little indication as to how there are reconciled or segregated. I will show some behavioral data suggesting dominance of familiarity in face stimuli, and that of novelty in natural scenes.

Close Window

Vision During Action (David Knill, Session Chair)

David Burr, University of Florence

Keeping vision stable

One of the more intriguing mysteries of visual perception is how in the face of continual saccadic eye movements do we construct a stable representation of the world based in external rather than retinal coordinates. This question is far from solved, but progress is being made. I will report a series of studies from our group, describing the transient changes in perception that occur at the time of saccades and speculating on how these may contribute to stability. Studies show that the magno-cellular system is selectively suppressed; that visual space undergoes a transient but drastic distortion at the time of saccades, selectively for verbal reports, not for blind pointing; and that there is a dramatic compression, and even an inversion, of perceived time. I will also discuss recent evidence from fMRI studies showing that a motion selective area in the human brain, MT complex, is spatially selective in spatiotopic rather than retinotopic coordinates.

Close Window

Carol Colby, University of Pittsburgh

Attention and active vision

Vision is an active process. Attention interacts with incoming sensory signals and information about intended actions to construct a stable spatial representation of salient objects and locations. At the single neuron level, memory traces of attended stimuli are remapped when a saccade occurs. Responses to remapped stimulus traces are found in parietal, frontal and extrastriate cortex. In functional imaging studies in humans, remapping is evident throughout visual and parietal cortex. Remapping of salient stimuli ensures that the contents of visual memory are aligned with the new eye position at the end of the saccade and may contribute to spatial stability.

What is the brain circuit that produces remapping? Is it a purely cortical phenomenon? We tested split brain monkeys in an eye movement task that required remapping of a memory trace from one hemifield to the other. Their performance was initially profoundly impaired but recovered quickly. This recovery of function tells us that subcortical as well as cortical pathways can contribute to the transfer of memory traces from one side of the brain to the other. In single neuron studies of split brain monkeys, remapped memory traces were still observed in parietal cortex. These findings indicate that a unified and stable representation of visual space is constructed by a redundant circuit with a remarkable capacity for reorganization.

Close Window

Tirin Moore, Stanford University

Mechanisms of saccadic suppression in ventral visual cortex

Saccadic eye movements rapidly displace the image of the world on the retina several times per second. Our failure to notice these displacements, and instead to perceive the world as stable, is largely due to saccadic suppression, in which visual sensitivity is momentarily disrupted shortly before, during, and after each movement. Because saccadic suppression primarily affects motion and luminance contrast sensitivity, it is typically attributed to a disruption of the magnocellular pathway and of dorsal visual areas. Using a procedure based on psychophysical experiments, we have measured the dynamics of contrast sensitivity of neurons in ventral visual area V4 just prior to saccades. We find that the sensitivity of V4 neurons to luminance contrast is profoundly suppressed at the time of saccades. However, as with psychophysically measured suppression, the sensitivity of V4 neurons to chromatic contrast remains largely intact. Thus the luminance-selective feature of saccadic suppression is exhibited by individual neurons in this area. The results demonstrate that although the perceptual phenomenon of saccadic suppression is confined to particular stimulus parameters its underlying neural mechanism is not be confined to a particular visual pathway.

Close Window

Randolph Blake, Vanderbilt University

Actions can arbitrate visual conflict

Action and visual perception are inextricably linked: vision guides action and action can influence visual perception. In a series of studies, my colleagues and I have examined the influence of motor planning and execution in situations where one is faced with conflicting interpretations about the nature of a visual object located in a given region of visual space. To induce visual confusion, one set of studies used an ambiguous structure from motion sequence and another set of studies used dissimilar monocular images that instigated binocular rivalry. In both studies, conflict resolution was influenced by an observer’s motor behavior in relation to the stimuli being viewed. Remarkably, this influence of action on the resolution of visual conflict happened even when the object being controlled fell outside of visual awareness. Currently in progress are experiments examining the role of learned associations between motor responses and visual objects, including objects presented outside of awareness. Results from this work to date lead to speculations about the neural mechanisms underlying action’s influence on resolution of visual conflict. Supported by NIH EY13358.

Close Window

top