Combining information from different sensory modalities is important for accurate perception. Despite the ubiquity of this phenomenon, its neural basis is still poorly understood. Psychophysical studies have shown that integration of two cues from different modalities is, in many cases, statistically optimal. Using the illustrative case of auditory-visual spatial localization, we will address two key aspects of optimal multisensory perception. First, we show how optimality of cue integration can be implemented by simple linear operations on population patterns of activity, provided that neural variability belongs to a particular, broad family of distributions. This predicts additivity in bimodal interactions, which is consistent with physiological measurements. Second, a major limitation of current models of cue combination is that they assume that cues arise from a common source, which is often not the case. We present a Bayesian model that incorporates this ambiguity about the existence of a common source. This model accurately predicts human perception in an auditory-visual spatial localization task in which integration is not mandatory. Together, these findings are steps towards a complete model of multisensory perception.

Two studies are presented. First, we investigated the neural signals underlying the effects of a probabilistic spatial cue on accuracy in a motion discrimination task. We found that an extensive network of areas is involved in mediating the effects of the cue and that cue evoked BOLD signals that precede the stimulus presentation and the subject's decision vary on a trial by trial basis and predict subjects' performance with 75% accuracy. Secondly, I present the results of a longitudinal fMRI study of the neural basis of neglect and recovery from neglect. It is found that shortly after a right hemisphere stroke associated with neglect, dorsal parietal regions in the left hemisphere show large responses to visual targets. The magnitude of this response is however diminished in the same subjects following recovery. It is suggested that neglect may result from a dynamic imbalance between left and right hemispheres.

Currently, we know that simple visual stimuli evoke large but structured patterns of population activity in early areas of visual cortex. This is due in part to the broad tuning of individual cells for features such as orientation and position of the stimulus, and in part to the smooth mapping of each feature across the cortical surface. For example, an elongated line stimulus will evoke a broad line of activity in primary visual cortex that is approximately 1 mm in width. Within this general region, the highest activity will be found in particular patches or columns of cells with the appropriate orientation preference. Small changes in the position or orientation of the stimulus lead to overlapping but distinctly shifted patterns of population activity. Similar types of population responses have been found for other types of stimuli, and in other cortical areas. However, we still know relatively little about which how these large population responses are used to guide behavior. In my talk, I will present evidence indicating that monkeys use a broad but tuned readout of the population activity in the middle temporal area (MT) to guide their behavior in a motion detection task. This readout could be achieved by applying different weightings to different direction columns in MT.

When subjects view an ambiguous stimulus, perception alternates between several interpretations of it. Using several bi-stable visual displays, we show that the alternation rate between dominance of the two percepts reaches a maximum when the stimulus is maximally ambiguous. Maximum ambiguity is defined by a configuration of the stimulus in which both percepts are equally likely, and therefore they dominate in mean for equal amounts of times. For configurations of the stimulus outside this 'equidominance' point (that is, when one percept dominates more than other), the alternation rate decreases drastically. Surprisingly, it is found that the alternation rate correlates with the entropy of a stochastic binary system in which state probabilities are defined in terms of the fraction of dominance of each percept. Furthermore, we present simple energy-based models as well as rate-based attractor models with input divisive normalization that can account for this behavior, providing a conceptual framework to interpret these psychophysical data in bi-stable perception.

One of the most important tasks accomplished by the visual system is piecing together a visual scene that spans the receptive fields of many neurons. Our brain is able to construct a coherent picture of our environment from neurons that take small spatially discrete samples of temporally discrete events. We have explored the mechanisms of global form processing by studying two aspects of neuronal function. First, at the level of single neurons, we have studied how spatial and temporal interactions among visual stimuli affect a neuron's response. I will describe how we have utilized sinusoidal gratings and perceptual pop-out stimuli to study the dynamics of suppressive influences both within and outside the classical receptive field of neurons in primary visual cortex. This work has helped to elucidate the structure of neuronal circuitry underlying form integration. Second, in neuronal ensembles, we have begun to explore how form processing is accomplished by a distributed network of neurons. I will present our work using a 100-electrode array to record from ensembles of neurons in primary visual cortex. This method has provided a powerful way to measure the dependence of correlation on stimulus properties and cortical distance.

Neurons in area MT are selective for the direction of visual motion. In addition, many MT cells are tuned for the global motion of a compound stimulus invariant of the motion components that comprise it, a behavior not seen in earlier visual areas. In this talk, I will demonstrate that this invariance can be captured by a model similar to the one proposed by Hubel and Wiesel to describe the construction of orientation tuning in V1. This "cascade" model describes an MT cell as summing the afferent responses of a population of nonlinear V1 cells, followed by a simple nonlinearity to capture thresholding. Fits of the model show that it robustly predicts the separately measured responses to gratings and plaids. The model also captures the full range of invariance across MT cells. Invariant cells that signal pattern motion are distinguished by convergent excitatory input from V1 cells with a wide range of preferred directions, strong motion opponent suppression and a tuned normalization that may reflect suppressive input from the surround of V1 cells.

The lecture will describe the recovery of the overall visual system after exposure of the eye to very intense "bleaching" exposures, and the cellular and molecular mechanisms involved in the regeneration of rhodopsin. A model is presented for the delivery of 11-cis retinal to opsin in the bleached rods, which accounts for the kinetics of psychophysical dark adaptation and rhodopsin regeneration, both in normal subjects and in a number of disease conditions.

Human visual motion perception is qualitatively consistent with a Bayesian observer model that optimally combines noisy sensory information with a prior preference for slow motions. However, a quantitative validation of this concept is difficult because the noise characteristics and the prior expectations are not directly measurable. I will present an extended Bayesian observer model that not only accounts for the bias but also the trial-by-trial variability of human subjects performing a visual speed discrimination task. By fitting the model directly to psychophysical data, the prior expectations and noise characteristics of each subject become uniquely constrained and can be extracted. This quantitative characterization of the Bayesian observer model not only allows us to test the accuracy of human's prior assumptions about the observed visual motion, but also to constrain potential neural instantiations of the model. I will discuss ongoing experimental and modeling work aimed at linking neural population responses in area MT to perception and the model. Finally, I will demonstrate the generality of the Bayesian observer model by showing how it can account for the effects of sensory adaptation.

Neurophysiologists have amply documented the existence of center-surround interactions in motion processing. Proposed functional roles of this ubiquitous neural mechanism include figure-ground segregation and the analysis of object shape, but until recently those putative roles were untested conjectures. In a series of experiments, we investigated the properties of center-surround interactions: from a behavioral description and to a functional characterization.

Several years ago we demonstrated that increasing the size of a high-contrast moving object can make its motion substantially more difficult to perceive (Tadin et al., Nature, 424, 312-5, 2003). Based on converging lines of evidence, we attributed this counterintuitive finding to the involvement of motion-sensitive neurons with suppressive center-surround receptive fields, such as those found in area MT. In subsequent studies, we explored center-surround interactions using a variety of psychophysical methods and phenomena, including threshold measurements, reaction times, motion-aftereffect, binocular rivalry and modeling. Moreover, using a reverse correlation approach, we documented fine temporal delays associated with the interaction between center and surround signals.

What are functional roles of center-surround suppression in motion? Special population studies revealed that center-surround suppression is weaker in elderly and patients with schizophrenia - a result responsible for their paradoxically better-than-normal performance in some conditions. Moreover, these two subject groups exhibit deficits in figure-ground segregation, suggesting a possible functional connection. In a recent study, we directly addressed this possibility and report experimental evidence for a functional link between center-surround suppression and figure-ground segregation.

Humans regularly make inferences that go beyond the data they have seen. Two questions immediately arise: what is the knowledge that supports these inferences, and how is this knowledge acquired? I will present a hierarchical Bayesian approach to inductive reasoning that addresses both questions. When making inferences about the distribution of a novel property, people draw on rich semantic knowledge that can often be captured using structured representations of the relationships between the entities in a domain. For instance, given that gazelles have T4 cells and carry E. spirus bacteria, taxonomic relations are useful for predicting which other animals are likely to have T4 cells, but predator-prey relations are more useful when reasoning about the distribution of E. spirus bacteria. I will show that our formal framework provides close quantitative fits to human inferences about several kinds of properties when supplied with appropriate knowledge representations for each task. Different inductive tasks often draw on different kinds of knowledge which are best captured by qualitatively different kinds of representations. For instance, anatomical features of biological species are best captured by a taxonomic tree, political views are best captured by a linear spectrum, and friendship relations are best captured by a set of discrete cliques. Our hierarchical framework helps to explain how humans can discover the best kind of representation for a given inductive context.

Human depth perception involves combining multiple, possibly conflicting, sensory measurements. Previous work with slightly conflicting cues has shown that this process is performed by statistical optimal weighted averaging. Here we ask whether the brain has a mechanism to be robust to large cue conflicts. We investigated how disparity and texture are combined in estimating slant as a function of their conflict. When the two cues only had a small conflict, we found evidence for optimally weighted averaging. At larger conflicts, we observed robust behavior in which one of the discrepant cues was rejected. Interestingly, the ignored cue could be either disparity or texture, and was not necessarily the less reliable cue. Optimally weighted averaging has previously been modeled as the combination of Gaussian sensory estimates. We show that both weighted averaging and robustness are predicted if the tails of the sensory estimates are heavier than a Gaussian. Lastly, we probed to see whether access to single-cue estimates determined robustness behavior. We found no evidence for access, suggesting nearly full cue fusion. We used this data to estimate a 'coupling prior' for disparity-texture combination.

A large body of data suggests that top-down influences from higher-level visual areas affect response properties of neurons in lower-level visual areas. In this talk, I will describe two models to explain how and why the brain uses these top-down interactions in visual processing.

The first model proposes that changes in early-level response properties result from redundancy reduction mechanisms. Since natural images are highly correlated in space and time and the brain’s resources are limited, top-down mechanisms work to avoid representing this redundant information. I will illustrate the model using data from cat LGN and V1.

The second model proposes that initial feedforward processing to higher-level visual areas is fast but coarse and does not capture all the relevant information in the input. Subsequent feedback mechanisms highlight high-resolution information in lower-level visual areas for further processing. These mechanisms provide a top-down interpretation of the input to lower-level areas and seem related to attentional mechanisms. I will illustrate this framework using data from neurons in monkey visual cortex.

Primary visual cortex (V1) contains a well-ordered topographic map of the entire contralateral visual hemifield. In macaque monkey the mapping between visual space and V1 is well described by complex-logarithm models, which require only a few parameters to account for the two-dimensional structure of the mapping, including cortical magnification and the slight local anisotropy in the visual field representation. I will describe an fMRI experiment which establishes that complex-logarithm models generalize to human visual cortex and that there is little intersubject variability in V1 topography. Because stimulating the foveal and peripheral visual field in the MRI scanner is challenging, accurately measuring the topographic map in these regions is not yet practical. However, measuring the shape of V1 can provide information about topography over the entire cortical area. I will present the results of a study where V1 was imaged using high-resolution structural MRI of the stria of Gennari at 7 T in whole ex vivo human hemispheres. A surface mesh representation of V1 was constructed from the MRI data then accurately flattened into the plane to allow statistical shape analysis. V1 shape was found to be nearly invariant across subjects, which is consistent with the results of the in vivo fMRI experiment. Also, the measured shape of V1 was in excellent agreement with the shape predicted by the model of topography. The results of both these studies indicate regularity in the topography of V1 across humans and macaques.

Nicotinamide adenine dinucleotide (NADH) is the principle electron carrier in glycolytic and oxidative metabolism. Importantly, the reduced co-enzyme (NADH) is fluorescent, while the oxidized co-enzyme (NAD+) is not. Therefore, intrinsic NADH fluorescence gives a direct measure of the cellular NADH/NAD+-ratio and has been utilized as an indicator for both oxidative and glycolytic metabolism.

The principle of two-photon NADH imaging (Kasischke et al., 2004) is that a native intracellular molecule present in millimolar concentrations is detected in live animals or tissues using non-linear laser-scanning microscopies. Advantages over established functional neuroimaging modalities are the absence of dyes and tracers, spatial resolution ranging from the columnar to the subcellular level, and temporal resolution below the response functions of glycolytic or oxidative metabolism and cerebral blood flow. In addition, because of its unique molecular structure and its specific binding to cytoplasmic and mitochondrial dehydrogenases, NADH is a probe of the cellular microenvironment readily accessible to steady-state and time-resolved spectroscopy and possibly imaging spectroscopy (Vishwasrao et al., 2005).

Our latest results show that we can detect activity-dependent physiological metabolic fluctuations within sharply defined cortical areas and possibly even within single cells. Under pathological conditions, such as cerebral hypoxia, we obtain a precise spatiotemporal representation of biochemical disturbances in the mouse cortex. These disturbances are highly dependent on their location within the capillary bed.

The application of two-photon NADH imaging bears a great potential as a novel microscopic functional neuroimaging technique. Possible extensions of this imaging modality are time-resolved imaging spectroscopy of intracellular conformational changes of NADH and visualization of oxygen diffusion in the brain.

The ability to image single cells noninvasively in the living retina has important applications for the study of normal retina, diseased retina, and the efficacy of therapies for retinal disease. In this thesis I describe a new instrument for high resolution, in vivo imaging of the mammalian retina that combines the benefits of confocal detection, adaptive optics, multispectral, and fluorescence imaging. The instrument is capable of imaging single ganglion cells and sub-cellular structures following retrograde transport into ganglion cells of fluorescent dyes injected into the monkey lateral geniculate nucleus (LGN). By comparing the dendritic structures of imaged ganglion cells, different types of ganglion cells can be classified. The resolution of the instrument has been demonstrated by imaging the sub-micron structures of ganglion cells. In addition, we developed a new imaging method that involves simultaneous imaging in two spectral bands, thus allowing the integration of very weak signals across many frames despite inter-frame movement of the eye. With this method, we have been able to resolve the smallest retinal capillaries using fluorescein angiography, as well as the mosaic of retinal pigment epithelial (RPE) cells revealed by lipofuscin autofluorescence.

The current approach to the problem of cue integration in the perception of 3D shape postulates that the outputs of (fairly) independent depth-processing modules are combined in a statistically optimal fashion. An alternative to the modular architecture of the Modified Weak Fusion Model has been proposed by Domini and Caudek (2006) by hypothesizing that the visual system combines the information provided by different image signals (e.g. velocities, disparities, texture gradients ...) prior to the extraction of depth information. Such approach has been termed the Intrinsic Constraint Model. The aim of this presentation is twofold. First, I will discuss novel empirical findings about perception of structure from stereo and motion signals that are consistent with the predictions of the IC model. Second, I will carry out an ideal observer analysis showing that the IC model yields optimal performance for the recovery of the affine structure.

Experimental evidence suggests that communication within the first layer of the visual cortex is best modeled using a geometry where not all directions are treated equally. A model proposed by Citti and Sarti identifies the visual cortex as a 3D manifold known as the rototranslation space. A visual image is identified as a surface within this manifold. We'll discuss some predictions based on the mathematics of this model for how the brain interprets visual information. In particular, we'll look at the image completion or in-filling problem. Mathematically this is very similar to the classical problem of finding surfaces with minimal area. There are explicit partial solutions which should yield insight into the issue of image completion.

There are two extreme views regarding the source of variability in the response of cortical neurons. According to the optimistic view, the brain is deterministic and the variability simply reflects internal variables not under experimental control (e.g. expectations, desires, attention, etc). By contrast, the pessimistic view asserts that all of the variability is due to noise internal to the nervous system. Unfortunately, neither of these views can be reconciled with ideal observer analysis whether at the behavioral and neural level. I will suggest an alternative according to which the variability is caused by two major sources which have nothing to do with internal variables or the stochasticity of neural hardware. These sources are the ill-posed nature of computations faced by the brain, and the fact that cortical computations are necessarily suboptimal. By suboptimal, I mean that the brain only has partial knowledge of the statistical process that give rise to sensory inputs. Thus, according to this view, ignorance is the main cause of cortical variability.

Understanding the validity of functional magnetic resonance imaging (fMRI) blood oxygen level dependent (BOLD) contrast as an indicator of the functional status of brain tissue in the visual cortex is important for both theoretical and clinical purposes. One method to determine the validity is to compare an fMRI-based visual field map with behavioral perimetry. Ideally, these data should be obtained simultaneously to ensure that both measures are closely correlated. To facilitate this goal, a Video Automated Perimeter (VAP) was developed to facilitate visual field testing inside the MRI scanner. To confirm the VAP was comparable with standard automatic perimetry, data were obtained using both the VAP and a conventional Humphrey Field Analyzer (HFA) for 9 patients having stable, localized, cortical visual field defects. The spatial pattern of visual sensitivity matched well in the two tests, with a median spatial cross correlation of 0.71. Thus the VAP system provides visual field maps equivalent to those obtained with HFA, but can be administered while the patient is in the MRI scanner, thus reducing test time and ensuring that both measures are obtained under closely matched conditions.

VAP was then used to test the overall validity of fMRI visual field mapping. Forty-eight patients with cortical visual field loss participated in a total of 68 mapping studies. The results of these studies were analyzed, using both automated and manual scoring techniques to compare behavioral sensitivity and visually evoked fMRI responses for individual points and for an array of sectors segmenting out to 24B eccentricity. Overall, the two measures concurred in 68% of the locations tested. However, many of the mismatched points were diagnostically informative. Mismatches where there is an activation loss in the FFMap but vision is preserved in perimetry may represent neurovascular uncoupling caused by direct effects of cortical pathology on the vasculature and BOLD mechanism. Alternatively, mismatches when fMRI activation is preserved and perimetry shows an apparent loss of vision can arise if sensory input to early visual cortex is preserved but higher level processing is impaired, as in attentional neglect. Appreciating the validity of fMRI activation as an indicator of cortical function provides an important scale for weighing theoretical interpretations or clinical decisions, especially when misinterpretation could have important consequences for the patient, such as surgery-induced vision loss.

Action video game players (VGPs) have been shown to outperform their non-game playing (NVGPs) peers on a number of sensory/cognitive measures. This performance difference has been most consistently manifested as a large decrease in reaction time (RT) in VGPs compared to NVGPs. However, there are many possible mechanistic explanations for a reduction in RT (increased sensitivity to the stimulus, reduced criteria, faster motor execution) that cannot be easily teased apart by examining the research to date. Here we present data from two sensory integration tasks, a standard motion coherence paradigm (Newsome et al, 1989) and a novel auditory localization task, which in combination with a model developed by Palmer et al (2005), allows for a more explicit test of the relative contribution of sensitivity, criteria, and motor execution in generating the differences observed between VGPs and NVGPs. In both the motion and auditory tasks, VGPs demonstrated a large reduction in RT compared to NVGPs with equivalent accuracy. This pattern was well captured by the model with an increase in the rate of information accrual as well as a decrease in criteria in the VGPs. A set of follow-up experiments adds further support to the hypothesis that VGPs acquire sensory information more rapidly than NVGPs.

Visual scientists have shown that people are capable of perceptual learning in a large variety of circumstances. Nonetheless, the mechanisms mediating such learning are poorly understood. How flexible are these mechanisms? How are they constrained? We investigated these questions in two studies of perceptual learning. In both studies, we modeled subjects as observers performing probabilistic perceptual inferences to determine how their use of the available sensory information changed as a result of training. The first study consisted of five experiments examining the mechanisms of perceptual cue acquisition. Subjects were placed in novel environments containing systematic statistical relationships among scene and perceptual variables. These relationships could be either consistent or inconsistent with the types of sensory relationships that occur in natural environments. We found that subjects' learning was biased to favor statistical relationships consistent with those found in natural environments and proposed a new constraint on early perceptual learning to account for these results, defined in terms of Bayesian networks. The second study examined the mechanisms of learning in image-based perceptual discrimination tasks. Previous studies have demonstrated that people can integrate information from multiple perceptual cues in a statistically optimal manner when judging properties of surfaces in a scene. We wanted to determine whether subjects can learn to integrate optimally across arbitrary low-level visual features when making image-based discriminations. To investigate this question, we developed a novel and efficient modification of the classification image technique and conducted two experiments that explored subjects' discrimination strategies using this improved technique. We found that, with practice, subjects modified their decision strategies in a manner consistent with optimal feature combination, giving greater weight to reliable features and less weight to unreliable features. Thus, just as researchers have previously demonstrated that people are sensitive to the reliabilities of conventionally-defined cues when judging the depth or slant of a surface, we demonstrate that they are likewise sensitive to the reliabilities of arbitrary low-level features when making learning to make image-based discriminations.

The current study is designed to investigate the extent of preservation of three-dimensional (3D) visual spatial information across saccades and the role of eye- and head-movement (extraretinal) signals in transsaccadic integration. Specifically, we tested the following hypotheses: (i) visual spatial information within a three-dimensional environment is preserved across saccades; and (ii) information about eye and head position/movements contributes to the transsaccadic integration process. In order to precisely control the amount of visual information available to the participants within a single fixation and at the same time present visual information from a 3D environment, we conducted the experiments using an immersive virtual environment. We designed a spatial localization task that was modeled on Hayhoe & Lachter (Hayhoe and Lachter 1991), modified to include 3D visual information rather than 2D and presented within a controlled virtual environment rather than in the dark. In Experiment 1, we measured performance in conditions where it was possible for all the task-specific components to be seen simultaneously so that transsaccadic integration was not required to perform the task and compared it to performance in conditions where it was impossible for all the task-specific components to be seen simultaneously and transsaccadic integration was required. Performance decreased significantly when all task-specific components were not simultaneously visible. In Experiment 2, we explored the effect of systematically sampling and presenting pieces of visual information and found that decreasing the distance between successive views, and consequently increasing the amount of overlap between views, significantly improved performance in the spatial localization task.

The American National Standards Institute (ANSI) provides a standard for the safe use of lasers from which maximum permissible exposures (MPE) of laser light can be calculated. This standard is used as a limit for exposures of light to the eye and to assess the safety of retinal imaging experiments. Here we detect retinal changes after visible (568 nm) light exposures by using adaptive optics (AO) scanning laser ophthalmoscopy to image the cones and retinal pigment epithelial (RPE) cells in the macaque retina in vivo.

We compared the retinal changes caused by an extended uniform field with a field delivered by a scanning beam, both with and without AO. With equivalent average power these exposures showed similar immediate and long-term retinal changes. The retinal changes are likely a result of the average exposure power rather than the light delivery method. We also performed several infrared (830 nm) light exposures at levels near the ANSI standard. These showed neither decreases in autofluorescence intensity nor long-term retinal changes.

We have observed retinal damage in the macaque after exposure to light levels that are at or below the ANSI standard using AOSLO imaging of the cones and RPE cells. Within a week this damage also becomes visible in standard fundus photographs and fluorescein angiography. These results emphasize the need to minimize light exposures in retinal imaging and in ophthalmic procedures including intraocular surgery and photodynamic therapy.

The human visual system organizes motion into coherent percepts. The Gestalt psychologists described the visual system as preferring perceptions with "Pragnanz", or figural goodness. These phenomena are now understood in terms of Occam's Razor, or the preference for simple explanations. A great deal of work has gone into investigating how the visual system selects the best percept for a given motion stimulus. Meanwhile, much less work has looked at how the visual system decides what kind of moving objects to consider. These two problems are called parameter estimation, and model selection.

A simple motion estimation task with a limited range of plausible models acts as a probe into how the human visual system selects between models for stimuli. Two models, a rolling wheel and a bouncing stick take a synchronously moving two-dimensional 2-point-light display as input. The input, consistent with both a rolling wheel and a bouncing stick, was evaluated by both models. The relative probabilities of the models were consistent with earlier psychophysical work by Proffitt, Cutting, and Stier (1979).

In guinea pig retina immunostaining reveals a dual gradient of opsins: cones expressing opsin sensitive to medium wavelengths (M) predominate in the upper retina, whereas cones expressing opsin sensitive to shorter wavelengths (S) predominate in the lower retina. Whether these gradients correspond to functional gradients in post-receptoral neurons along both achromatic and chromatic retinal pathways is largely unknown? For achromatic retinal pathway, using monochromatic flashes, we measured the relative weights with which M, S, and rod signals contribute to horizontal cell responses. Using flickering stimuli produced by various mixtures of blue and green primary lights, we measured the relative weights of brisk-transient ganglion cells. We found that achromatic retinal pathway preserved the dual gradient (Yin et al., 2006). For chromatic retinal pathway, using monochromatic flashes, we measured the relative weights of both S+/M- and S-/M+ opponent ganglion cells and revealed the dendritic morphologies of those rare cells. We found that chromatic retinal pathway followed the dual gradient, but this led to minimal opponency in opponent ganglion cells in inferior retina. We think that the dual gradient might indeed be an evolutionary adaptation for achromatic vision, which trades off with chromatic vision.

The role of the striate cortex (area V1) in shaping our perception of visual patterns and scenes is of fundamental importance in understanding how we see. Visual suppression, where a salient stimulus is seen to temporarily disappear, provides a paradigm by which neuroscientists can isolate and study neural processes that are directly involved in perception. This approach has been applied to study electrophysiological activity in monkeys, as well as fMRI activation patterns in humans. Interestingly, the different types of studies have provided divergent results regarding the role of V1 in perceptual processing. Human neuroimaging (fMRI) studies have repeatedly shown that during perceptual suppression there is a marked decrease in the BOLD response in V1. In contrast, single unit studies in monkeys have reported the near absence of such suppression in the firing of individual V1 neurons. I will present new data in which we directly compare neurophysiological and fMRI responses during perceptual suppression in monkey area V1. By monitoring the two signals in a variety of conditions within the very same monkeys, we found that the single-unit and BOLD activation specifically ceased to be correlated during periods of perceptual suppression. Among the electrophysiological signals, only low frequency bands of the local field potential showed power decreases reflective of perceptual suppression, and these were restricted to the supragranular layers. I will discuss how the spatiotemporal pattern of neural events in V1 might contribute to shaping the BOLD response, and will argue that various signals provides a complementary, rather than contradictory, perspectives on neural processing within a cortical area.

We are constantly learning new things as we go about our lives. In addition to learning new facts, procedures and concepts, we are also refining our sensory abilities. How and when these sensory modifications take place is the focus of intense study and debate. While sensory improvements were thought only to occur when attention is focused on the stimuli to be learned (task-relevant learning), recent studies demonstrate performance improvements independent of the focus of attention (task-irrelevant learning). In this talk, I will discuss recent advances in our understanding of task-irrelevant learning and how this learning may take place through reward-related learning signals.

Separating objects from their backgrounds is a fundamental aspect of early visual processing and a pre-requisite for object recognition. I will describe a series of studies that use multi-input nonlinear analysis methods and high-density EEG recordings to track the temporal evolution of figure and background activity through several visual areas identified by fMRI mapping studies as well as in more anterior areas. In these studies the figure and background regions of simple texture-defined forms were tagged by distinct temporal frequencies or by independent m-sequences. Tagging allows us to reconstruct separate evoked responses from figure and background regions, even when they are presented simultaneously. The lateral occipital complex (LOC), identified on the basis of an intact vs scrambled object localizer, expressed a strong preference for the figure region, relative to the background, well before 100 msec. The figural response in the LOC is largely cue-invariant and shows a degree of position invariance as well. An enhanced figure response was observed in early retinotopic visual areas between 250 and 350 msec, consistent with feedback activation after segmentation by higher-order areas such as the LOC. Figural activity was also seen in anterior infero-temporal and frontal cortical areas as early as 100 msec. EEG source imaging, combined with multi-input tagging methods shows that figure ground segmentation involves a distributed processing network extending well beyond occipital cortex.

Abnormal visual experience has been demonstrated to disrupt the postnatal development of the visual system; both synaptic refinement in visual cortex and the growth of the eye. The information available to the neural visual system and the activity-dependent processes is defined by the retinal image formed in the infant eye. I will present our recent studies of the optical quality of the human infant eye (monochromatic and chromatic aberrations) and the degree to which the developing visual system controls its own postnatal visual experience through accommodation (sensitivity to defocus, dynamics and steady-state stability). I will discuss these studies in the context of the information available to guide normal and abnormal development.

Attention is generally viewed as a way to reduce the sensory load that the central perceptual system must analyze, and is measured by improvements in perception when attention is summoned to the location of the stimulus to be discriminated. However, attention can also filter the information that is sent from higher centers to the motor system. For example, the decision of whether or when to make a saccade depends on the cost vs. benefit of making the saccade, and this, in turn, depends on whether one is inspecting a fine feature of a large object thus requiring foveal examination, or viewing the object in its surroundings, thus gaining no benefit from small saccades. The motoric system may need only to know the size of the attentional field to make the optimal decision. I will discuss evidence in support of this point of view.

Given that saccades are tightly coupled to attention, it is perhaps not surprising that the oculomotor system can learn to adjust saccades to meet the visual needs of the moment, not simply to act as targeting machines. I will discuss experiments showing that saccadic adaptation has a flexibility far in excess of what is required to adjust for slow changes in the oculomotor apparatus.

We are interested in optical quality on the retina and early stage sampling of the optical image during emmetropization and myopia development. In chick we have shown that retinal image quality due to higher-order aberrations reduces during normal growth and that this reduction is interrupted by the lens induction of myopia. We have developed a new, simple metric of retinal image quality, equivalent retinal blur.

These results are suggestive of the emmetropization of the aberration component of retinal image quality in addition to refractive error. Our earlier results in fish indicated that the fine optical structure of the gradient refractive index (GRIN) within the crystalline lens could be influenced by visual function.

To further investigate whether changes in optical quality are indirectly a result of eye growth, we have developed eye models in which the optical elements scale uniformly with eye size. In a second set of models, we have held the linear blur on the retina constant as the eye grows.

Using published rates of ocular growth for chick, monkey and human, we predict the defocus, optical aberrations and image quality as a function of age. The comparison of model predictions with experimental measurements allows us to determine the relative changes on the retina of differing components of retinal image quality. These changes are species specific but are supportive of an emmetropization of higher order aberrations in all three species.

In vivo angular cone spacing and linear rod spacing measurements in the chick are however suggestive of uniform ocular expansion, with no improvement in retinal sampling with age. Literature ganglion cell density measurements are consistent with improved retinal sampling with age in chick. This in turn could underlie emmetropization of retinal blur.

In this talk, I will present a series of completed, pilot, and proposed experiments that (1) characterize center-surround interactions in motion perception and (2) link this basic contextual mechanism with our ability to segregate moving objects from the background.

First, I will present a multi-method approach designed to characterize center-surround interactions in motion by focusing on both the nature of the mechanism and its neural correlates. For example, a TMS pilot study showed that 15min of 1Hz TMS stimulation of cortical area MT substantially wakens center-surround suppression as indicated by a large reduction of thresholds for perceiving motion of large high-contrast objects. Other completed/planned studies are relying on reaction times, computational modeling (RT diffusion model), repetitive and single pulse TMS, elderly subjects as a model associated with reduced GABAergic function, and pharmacological manipulations of GABA in normal subjects, along with associated magnetic resonance spectroscopy measurements.

Second part of the talk will focus on linking center-surround mechanisms with its likely functional role in perception: ability to segregate objects from the background. Here I will rely on experimental manipulations of center-surround interactions within observers and on individual differences between observers to link changes in the strength of center-surround mechanisms with associated changes in figure-ground segregation ability. Other studies will exploit perceptual learning, RT modeling, TMS and pharmacological GABA manipulations to provide evidence for a functional connection between center-surround interactions and figure-ground segregation.