ABSTRACTS – 2016 AIC ================================================================================== Wyatt Bair University of Washington A binocular model for motion integration in MT neurons Visual neuroscience is a field of great specialization, where submodalities such as motion, depth, form and color are often studied in isolation. One disadvantage of this isolation is that results from each subfield are not integrated to constrain common underlying neural circuitry. Yet, to deeply understand the cortical computations underlying visual perception, it is important to unify our fragmentary models to reach a critical mass of constraints so that robust and powerful circuits can emerge. I will discuss our efforts to unify models of direction selectivity, binocular disparity and motion-in-depth (MID, or 3D motion) to reveal circuits and to understand computations from V1 to area MT and beyond. In spite of the great attention given to area MT in terms of its role in motion perception, past efforts to model this area have largely overlooked the binocular integration of motion signals. Recent electrophysiological studies have tested binocular integration in MT and found surprisingly that (i) MT neurons lose their hallmark pattern motion(cid:0) selectivity when stimuli are presented dichoptically and (ii) many MT neurons are selective for motion-in-depth (MID), contrary to the prevailing view. By unifying these novel observations with insights from monocular, frontoparallel motion studies concurrently in a binocular MT motion model, we have generated clear, testable predictions about the circuitry and mechanisms underlying visual motion processing. We built binocular models in which signals from left- and right-eye streams could be integrated at various stages from V1 to MT and attempted to create the simplest plausible circuits that accounted for the physiological data. Our successful models make predictions about the existence and the order of critical operations along the pathway leading to MT, and hint at unexpected relationships between pattern- and 3D-motion processing. I will present our results and insights and discuss the challenges of trying to unify constraints across visual sub-modalities to build more robust and comprehensive models of early- and mid-level visual processing. Dirk Bernhardt-Walther University of Toronto Contour junctions underlie neural representations of scene categories in human v Authors: Dirk Bernhardt-Walther and Heeyoung Choo Humans can quickly recognize a novel real-world environment according to its basic-level scene category. Which visual features facilitate such efficient categorical scene perception in humans? To answer this question we combine multi-voxel pattern analysis of neural representations of scene categories throughout visual cortex with computational models of scene categorization. Participants viewed line drawings of six scene categories (beaches, forests, mountains, city streets, highways, and offices) while in an MRI scanner. We decoded scene categories from locally distributed neural activity patterns. We compared decoding error patterns to the error patterns from five computational models of scenes, each relying on the statistics of only one kind of contour property: orientation, length, curvature, junction types, or junction angles. We found that statistics of junction properties exhibited the largest contribution to the neural representations of scene categories in high-level scene-selective brain regions (parahippocampal place areas, PPA; occipital place area, OPA; lateral occipital complex, LOC), followed by orientation statistics (see figure). To assess the causal involvement of these visual properties in neurally representing scene categories, we manipulated the images in such a way that junctions or orientations were disrupted. In early visual cortex, scene categories could be successfully decoded under both manipulations. In the PPA, OPA and LOC, on the other hand, disruption of junction statistics severely reduced the extent of category-specific neural activity. When orientations were disrupted, scene categories could still be decoded successfully in the PPA and LOC. Based on these results we suggest a causal role for contour junctions, which provide cues for the 3D arrangement of surfaces in a scene, in the neural mechanisms of scene categorization. 1 Vincent Billock The Ohio State University Visual Amplification Via Sensory Integration in Rattlesnakes, Cats and Humans Sensory integration and sensory binding are similar problems separated by a vast methodological gulf. The dominant paradigm of binding theory is neural synchronization, while sensory integration is built on observations of bimodal neurons. These cells show large increases in firing rates for bimodal presentation of weak stimuli, but little improvement for strong stimuli, a finding known as the Principle of Inverse Enhancement. It would be useful to link these two fields so that methods from each could be used by the other. The best case for such a bridge is the rattlesnake, which has two dissimilar visual systems, one for light and one for heat. Although this sounds like a binding problem, the rattlesnake has been studied using the methods of sensory integration. Many cells in rattlesnake optic tectum are sensitive only to light but can be strongly modulated by heat stimuli, or vice versa. I simulated these cells by assuming that they are members of synchronized pairs of excitatory-coupled rate-coded neurons. The same synchronized neuron model, without any parameter changes, accounts for a population of cells in cat visual cortex whose firing rates are enhanced by auditory stimuli consistent with the Principle of Inverse Enhancement. It is intriguing that the most important principle in sensory integration can be derived from binding theory. The same mechanism can be used to model within-vision nonlinear perceptual amplifications, such as those seen in binocular vision and color vision. Alyssa Brewer University of California Irvine Rod scotoma fMRI elucidates cortical rod pathways and lesion measurement concerns Are silencing, ectopic shifts, and receptive field scaling in cortical scotoma projection zones (SPZs) the result of long-term reorganization (plasticity) or short-term adaptation? Electrophysiological studies of SPZs following retinal lesions in animal models remain controversial, because they are unable to conclusively answer this question due to limitations of the methodology. Here we used functional MRI visual field mapping via population receptive field (pRF) modeling with moving bar stimuli under photopic and scotopic conditions to measure the effects of the rod scotoma in human early visual cortex. As a naturally-occurring central scotoma, it has a large cortical representation, is free of traumatic lesion complications, is completely reversible, and has not reorganized under normal conditions (but can, as seen in rod monochromats). We found that the pRFs overlapping the SPZ in V1, V2, V3, hV4, and VO-1 generally (1) reduced their BOLD signal coherence, and (2) shift ed their pRFs more eccentric, but (3) scaled their pRF sizes in variable ways. Thus, silencing, ectopic shifts, and pRF scaling in SPZs are not unique identifiers of cortical reorganization; rather, they can be the expected result of short-term adaptation. However, are there differences between rod and cone signals in V1, V2, V3, hV4, and VO-1? We did not find differences for all five maps in more peripheral eccentricities, outside of rod scotoma influence, neither in coherence, in eccentricity representation, nor pRF size. Thus, rod and cone signals appear to be processed similarly in cortex. ================================================================================== Bruce Brown Brigham Young University Whole-wave EEG analysis in the identification of neuropsychiatric illnesses Authors: Bruce Brown, Dawson Hedges, Jack Silcox An eigenvector-based approach is used to extract unique ideographic cognitive components for individual subjects from ERP data gathered during a cognitive task. These cognitive components appear to have impressive personal identification and diagnostic capabilities. For example, females are found to have a substantially different Sternberg memory search process than males (F[1,33] = 579.84, p<.0001, Rsquare=.946). In similar studies, the discrimination between clinically depressed males and healthy controls is strong (F[1,318]= 1802.31, p<.0001, Rsquare=.850), as is the discrimination between OCD males and healthy controls (F[1,318]= 1939.76, p<.0001, Rsquare=.916). In the 2 neurological area cognitive components discriminated well between mild AD subjects and healthy controls (Wilks''lambda = .4297, p<.0001, Rsquare=.5703), even when using a weak cognitive task (auditory oddball). Additional studies are in progress. ================================================================================== Patricia Cheng University of California, Los Angeles Analytic knowledge of causal invariance in rats Authors: Julia Schroeder, Aaron Blaisdell, Rui He, Jeffrey Bye, Patricia Cheng Abstract: none yet ================================================================================== Lawrence Cormack The University of Texas at Austin The perception of depth vs. frontoparallel motion probed with motor tracking. From traditional psychophysics, we know that stereoscopic acuity is exquisite, yet under some conditions depth motion is more difficult to see than frontoparallel motion, this despite the ecological importance of detecting approaching objects. Being fond of more naturalistic tasks, we wondered how observers would fare if asked to dynamically track objects moving in three dimensions. In the main experiment, subjects pointed with their index finger to track the center of a stereoscopic target as it moved in a 3-dimensional Brownian random walk. We measured tracking performance by calculating the cross-correlograms for each of the cardinal motion axes. Tracking performance was selectively impaired for depth motion compared to horizontal and vertical (frontoparallel) motion, by which we mean that the peak correlation between stimulus and response was both lower and occurred after a longer time delay. Moreover, this impairment was greater than was expected from the small relative size of the retinal signals arising from depth motion. Further experiments ruled out motor explanations (e.g., the notion that moving the hand along a depth axis is more difficult than moving it within a frontoparallel plane). In another experiment, observers tracked a disparity-defined target in a dynamic random element stereogram as it moved in a 3-dimensional Brownian random walk. When frontoparallel tracking required a disparity computation, it showed the same response delay observed in the original 3D tracking experiment – a fact we share while remaining agnostic about the order (in the Lu and Sperling sense) of stereomotion processing. Thus, despite the crucial importance of egocentric depth motion, its perception and the motor responses thereto are actually impaired relative to that of frontoparallel motion. ================================================================================== Sven Dickinson University of Toronto Model-Based Perceptual Grouping and Shape Abstraction Authors: Sven Dickinson and Pablo Sala For many object classes, shape is the most generic feature for object categorization. However, when a strong shape prior, i.e., a target object, is not available, domain independent, mid-level shape priors must play a critical role in not only grouping causally related features, but regularizing or abstracting them to yield higher-order shape features that support object categorization. In this talk, I will present a framework in which mid-level shape priors take the form of a vocabulary of simple, user-defined 2-D part models. From the vocabulary, we learn to not only group oversegmented regions into parts, but to abstract the shapes of the region groups, yielding a set of abstract part hypotheses. However, the process of shape abstraction can be thought of as a form of "controlled hallucination", 3 which comes at the cost of many competing 2-D part hypotheses. To improve part hypothesis precision, we assume that the 2-D parts represent the component faces of aspects that model a vocabulary of 3-D part models. We then exploit the relational structure (spatial context) of the faces encoded in the aspects, and again formulate hypothesis selection in a graph-theoretic, probabilistic framework. Finally, we introduce a technique that is able to recover the pose and shape of a volumetric part from a recovered aspect, yielding a framework that revisits the classical problem of recovering a set of qualitative 3-D volumetric parts from a single 2-D image. Barbara Dosher University of California, Irvine Perceptual Learning ================================================================================== Frank Durgin Swarthmore College The possible role of angular expansion in the misperception of large-scale space Authors: Frank Durgin and Zhi Li It has long been observed that perceptions of distances and slant in large-scale spaces (e.g., those affording locomotion) are typically distorted: Hills look much steeper than they are -- even when looking downhill; distances along the ground are foreshortened. Based on a series of studies of perceived visual direction in pitch and yaw, on parametric examination of slant estimates for large and small surfaces at near and far distances, and on a variety of perceptual measures of perceived distance and direction in large-scale space, we propose that the exaggeration of perceived angular deviations from horizontal/straight-ahead (typically with a gain of 1.5 in pitch) can parsimoniously account for a great deal of new and existing data on perceptual bias - including the large-scale horizontal-vertical illusion. We speculate that these angular exaggerations must have a functional basis and propose that orientation biases may reflect a coding scheme for retaining precision at the cost of accuracy. Such a trade-off could aid in calibrated action control, because the calibration of action control is limited by the precision of feedback guiding calibration processes rather than by the accuracy of estimation. ================================================================================== James Elder York University The Southampton-York Natural Scenes (SYNS) Dataset Authors: James Elder, Wendy Adams, Erich Graf, Alex Muryy, Arthur Lugitigheid The inference of 3D structure from 2D images is a central function of biological and machine vision systems. Since the problem is ill-posed, optimal inference depends upon knowing the joint statistics of 3D surfaces and 2D images. We have developed a new public dataset (syns.soton.ac.uk) that can be used to ground theories of human visual processing and machine vision algorithms in the ecological statistics underlying the problem. To fairly represent the diverse visual environments we experience, we randomly sampled scenes from 19 outdoor and 6 indoor categories across Hampshire, UK. Outdoor categories, identified by the UK Land Use dataset, include cropland, coastal dunes, woodlands, industrial estates, wetlands, residential areas, farms and orchards. Indoor categories include residential, theatres, cafes and offices. Each scene is represented by three types of co-registered data: (i) 3D point clouds spanning 360° x 135° captured by a laser rangefinder (LiDAR), (ii) High dynamic range images spanning 360° x 180° captured by a spherical camera and (iii) 18 Stereo image pairs, each spanning 35° x 24° and tiling a 360° horizontal panorama, captured by a custom-built high-resolution stereo rig, with camera separation matched to average human interpupillary distance. LiDAR data were analysed to determine the distribution of egocentric surface attitudes in outdoor environments. Surface normals were computed at each LiDAR point, using an adaptive scale selection method. Overall, the 4 distribution is dominated by the ground plane. To relate these natural scene statistics to human perception, we conducted psychophysical experiments to measure both discrimination and absolute judgments of the 3D orientation of real textured planar surface; together these allowed estimation of the observer’s underlying prior. We found that priors varied substantially across observers, with an overall bias toward surface normals lying in the vertical meridian. We discuss potential reasons for the variability across observers and divergence between ecological statistics and human perceptual judgments. ================================================================================== Gregory Francis Purdue University A small part of the Human Brain Project: Neural dynamics of visual segmentation The Human Brain Project is a European funded effort to build a "scaffold" model and simulation of the human brain by 2023. In its current phase the HBP is developing simulation and database technologies that will be used to achieve the long-term goal. Some of these technologies may be of interest to cognitive scientists, especially if they develop neural models of human behavior. I will briefly describe some of the relevant HBP technologies and share how they contributed to a neural network model of visual segmentation. I will also describe how the model explains complex effects of perceptual grouping for empirical data on visual crowding. Wilson Geisler University of Texas at Austin Measurements and models of detection in natural scenes Authors: Steve Sebastian, Jared Abrams, Wilson S. Geisler An ultimate goal of vision science is to measure and predict performance in visual tasks under natural conditions. Perhaps the most fundamental task is to detect target objects in the natural backgrounds that surround us. It is known from experiments with simple stimuli that the specific properties of a background (luminance, contrast, similarity to the target) have a strong influence on detectability. It is also know from experiments with simple stimuli that the uncertainty created by randomizing the amplitude and/or location of the target (“target uncertainty”), and randomizing properties of the background (“background uncertainty”) are additional factors influencing detectability. What is relatively unknown are how these known factors individually affect detection in natural scenes, and how these factors combine in affecting detection in natural scenes. We address these two questions using a direct experimental approach that is quite efficient and could be used to address similar questions for other natural tasks. A large collection of calibrated natural images is divided into millions of background patches that are sorted into narrow bins along dimensions of interest. In the present study, each bin corresponds to a particular narrow range of luminance, contrast, and target similarity. Detection performance is then measured in a sparse subset of bins spanning the entire space, with and without target and background uncertainty. We find that detection thresholds in natural backgrounds vary linearly along all three dimensions and that humans are remarkably unaffected by simultaneous background and amplitude uncertainty. We show that these results are predicted by a Bayesian signal- detection model (a generalized matched template model) derived from first principles. ================================================================================== Joseph Houpt Wright State University Revisiting Stereoscopic Disparity as a Feature in Visual Search Authors: Joseph W. Houpt, Leslie M. Blaha, Megan B. Morris, John P. McIntire With the recent explosion in the number of commercially available stereoscopic 3D displays, it has become financially feasible to use 3D in many operational environments. Almost certainly there will be a wrong way to exploit the technology in the sense that it leads to worse performance and outcomes than when employing standard 2D displays, but there is still a lot of potential for the technology to enhance capabilities. The focus of this talk will be on the use of stereo-3D displays for visual search tasks and, in particular, on the use of stereoscopic disparity as a 5 target feature. I will begin with the standard demonstration of a pop-out effect for 3D targets and an extended analysis using distributional level measures. Next, I will discuss a follow-up study, in which we explored the perceptual processing of targets based on a combination of shape and stereoscopic disparity. Using Systems Factorial Technology, we found that participants are faster at identifying combined shape depth targets than would be predicted by independent, parallel search and that most participants employed a coactive strategy (i.e., they searched for the target based on pooled shape depth information rather than treating each source of information separately). ================================================================================== Xiaoping Hu Emory University The effect of prenatal alcohol exposure on brain connectivity in adolescents Authors, Zhihao Lik Bing Ji, Claire Coles, Mary Elln Lynch, Xiaoping Hu We performed resting state connectivity analysis and structural connectivity analysis in data from a sample of 72 prenatally alcohol exposed (defined as 13.3 oz absolute alcohol/week) individuals (age: 13±3; 37 male and 35 female) and 72 matched controls (age: 13.5±3; 40 male an 32 female). Our analysis successfully identified 7 functional networks (default mode, left frontal-parietal, right frontal-parietal, primary motor, primary visual, extrastriate visual, and salient) using independent analysis, and identified significant (p<0.05 corrected) reduction of functional connectivity in the exposed cohort all networks except the salient network. In addition, structural connectivities in the primary and extrastriate visual regions were examined. The results showed significantly decreased connectivity in the connections between the left and right primary visual cortices (p<0.03, Tukey HSD corrected), as well as between the left primary visual cortex and left extrastriate area (p<0.019, Tukey HSD corrected) in the PAE group. These results indicate that prenatal alcohol exposure leads to reduction in both functional and structural connectivities. ================================================================================== Alan Johnston University of Nottingham The harmonic vector average: a new approach to the aperture problem A central problem for the visual system is how to compute the speed of objects as they move in the world. The motion processing system in monkey and man is arranged in a hierarchy of anatomically distinct brain areas containing neurones whose receptive fields increase in size, response selectivity and computational complexity with increase in rank. The evident variation across the hierarchy in the spatial range of motion analysis, as indicated by receptive field size, leads to some unavoidable computational problems. Initial local analysis is limited by the aperture problem, a result of which is that neurones typically signal motion orthogonal to contours. These essentially independent estimates need to be brought together through some computation that can deliver the global motion of an object as a whole. We can study motion integration experimentally using an array of Gabor elements (Gaussian windowed moving sine gratings). The local speeds of object contours vary systematically with the cosine of the angle between the component of the local velocity orthogonal to the contour and the global object motion direction. A spatial distributed array of Gabor elements whose speed depends on local spatial orientation in accordance with this pattern can appear to cohere and move as a single surface. A number of models have been proposed to explain how the visual system might achieve this result. If we assume a single rigid translation, the global direction of motion can be found from at least two elements with different orientations, using a strategy known as the intersection of constraints (IOC). This strategy has usually been contrasted with the vector average. However, the vector average over local vectors that vary in direction always provides an underestimate of the true global speed, and if we have a biased set of local motions with respect to the global motion, the global percept is shifted towards the average direction, which is inconsistent with the IOC strategy. We need to look for an approach that will resolve these problems. If we plot the normal component motion vectors in a velocity space they lie on a circle through the origin. This circle when inverted in the unit circle maps to a line, allowing a least square estimate of the IOC and an average 6 inverse velocity for a set of normal components. This average, inverted once more in the unit circle, is the harmonic vector average (HVA). The harmonic vector average provides the correct global speed and direction for an unbiased sample of local velocities with respect to the global motion direction. The HVA over biased samples provides an aggregate velocity estimate that can still be combined through an IOC computation to give an accurate estimate of the global velocity, which is not true of the vector average. Psychophysical results for a biased distribution of Gabor arrays show perceived direction and speed falls close to the intersection of constraints direction for Gabor arrays having a wide range of orientations but the IOC prediction fails as the mean orientation shifts away from the global motion direction and the orientation range narrows. In this case perceived velocity generally defaults to the harmonic vector average. Neither the IOC nor the HVA can account for human global motion perception in biased arrays, however the perceived direction of motion appears to be bounded by the IOC and the HVA. ================================================================================== Philip Kellman University of California, Los Angeles Title: Spatiotemporal Boundary Formation Authors: Philip Kellman and Gennady Erlikhman Spatiotemporal boundary formation (SBF) is the perception of illusory contours, global form, and global motion from spatially and temporally sparse transformations of texture elements. Because it produces complete contours from elements lying along as little as 3% of an object's boundaries, SBF may be the "most from the least" in perceptual organization. In this talk, I consider recent progress in understanding how SBF works. Evidence suggests that local oriented edge fragments are somehow generated from discrete element changes. These fragments then connect to form continuous object boundaries through well-known contour interpolation processes. The mystery is how local oriented fragments are generated. Following formal proofs that local orientation could theoretically be derived from triplets of non- collinear element changes, we developed a paradigm for examining minimal conditions in SBF. We found that same display can appear as a single moving element along a sawtooth arrangement of dots or a larger oriented edge moving through the array, depending on timing. Experimental results indicating precise temporal constraints on SBF and the operation of edge formation in parallel across the visual field suggests that the local edge formation stage in SBF may depend on known spatiotemporal filter mechanisms (e.g., Adelson & Bergen, 1985; van Santen & Sperling, 1984). These “motion energy" filters are ordinarily studied with spatial orientation given unambiguously by luminance contrast, but our results suggest a duality whereby, when orientation is not specified by static information these motion filters are also spatiotemporal orientation detectors. A combination of known contour interpolation processes in middle vision and outputs of early spatiotemporal filters may explain the processes that produce shape and continuous boundaries in SBF. Michael Landy New York University Criterion learning in static and dynamic environments Authors: Michael S. Landy, Elyse H. Norton, Stephen M. Fleming, Nathaniel D. Daw Humans often make decisions based on uncertain sensory information. Signal detection theory describes detection and discrimination decisions as a comparison of stimulus strength to a fixed decision criterion. How is the criterion set? Here, we examine how observers learn to set a decision criterion in an orientation-discrimination task under both static and dynamic conditions. To investigate mechanisms underlying trial-by-trial criterion placement we compared covert and overt discrimination tasks. In each task, stimuli were ellipses with principle orientations drawn from two categories: Gaussian distributions with different means and equal variance. In the covert-criterion task, observers categorized a displayed ellipse. In the overt-criterion task, on every trial observers adjusted the orientation of a line that served as the discrimination criterion for a subsequently presented ellipse. We compared performance to the ideal Bayesian model and several suboptimal models that varied in both computational and memory demands. Under static and dynamic conditions, we found that, in both tasks, observers used suboptimal learning rules. A model in which the recent history of past samples determines a belief about category means fit the data best for most 7 observers and on average. Our results reveal dynamic adjustment of discrimination criterion, even after prolonged training. ================================================================================== Zhong-Lin Lu The Ohio State University qPR: An adaptive partial report procedure based on Bayesian Inference Authors: Zhong-Lin Lu, Jongsoo Baek, Luis Lesmes Iconic memory is best assessed with the partial report procedure, in which an array of letters appears briefly on the screen and a post-stimulus cue directs the observer to report the identity of the cued letter(s). Typically 6-8 cue delays or 600-800 trials are tested to measure the sensory memory decay function. Here we develop a quick partial report or qPR procedure based on a Bayesian adaptive framework to directly estimate the parameters of the sensory memory decay function with much reduced testing time. The exponential decay function is characterized by three parameters with a joint probability distribution. Starting with a prior of the parameters, the method selects the stimulus to maximize the expected information gain in the next test trial. It then updates the probability distribution of the parameters based on the observer’s response using Bayesian inference. The procedure is re-iterated until either the total number of trials or the precision of the parameter estimates reaches a certain criterion. Simulation studies showed that only 100 trials were necessary to reach an accuracy of ±2.5% correct and precision of 7.5%. A psychophysical validation experiment showed that estimates of the sensory memory decay function obtained with 100 qPR trials exhibited good precision (the half width of the 67% credible interval = 5.1%) and excellent agreement with those obtained with 1600 trials of the conventional procedure (mean RMSE = 5.7%). qPR relieves the data collection burden in characterizing sensory memory and makes it possible to assess sensory memory in clinical populations. Laurence Maloney New York University Representing and distorting probability and probability density Authors: Laurence T Maloney and Hang Zhang The movement we execute is not always the movement we plan; what we see is not always what is there; events in the world may turn out other than expected. Bayesian decision theory prescribes how to act so as to maximize expected value despite uncertainty concerning the outcomes of our actions. While human performance is impressive in many decision tasks, it is not optimal (Maloney, 2002). Small, systematic deviations from optimality are potentially a valuable source of information concerning how humans distort probability and probability density (Maloney & Mamassian, Visual Neurosci., 2009; Ting et al, J. Neurosci., 2015; Wu et al, PNAS, 2009, J. Neurosci., 2011; Zhang et al, PLoS Comp. Biol., 2010; Zhang & Maloney, 2012; Zhang et al, Frontiers in Neurosci., 2015). Based on this recent experimental work, I’ll outline an alternative to Bayesian decision theory based on more accurate characterizations of the human representation of probability and probability density and discuss why we distort probability and probability density as we do. ================================================================================== Jeff Mulligan NASA Ames Research Center Measuring and Modeling Shared Visual Attention Authors: Jeff Mulligan and Patrick Gontar Multi-person teams are sometimes responsible for critical tasks, such as flying an airliner. Here we present a method using gaze tracking data to assess shared visual attention, a term we use to describe the situation where team members are attending to a common set of elements in the environment. Gaze data are quantized with respect to a set of N areas of interest (AOIs); these are then used to construct a time series of N dimensional vectors, with each 8 vector component representing one of the AOIs, all set to 0 except for the component corresponding to the currently fixated AOI, which is set to 1. The resulting sequence of vectors can be averaged in time, with the result that each vector component represents the proportion of time that the corresponding AOI was fixated within the given time interval. We present two methods for comparing sequences of this sort, one based on computing the time- varying correlation of the averaged vectors, and another based on a chi-square test testing the hypothesis that the observed gaze proportions are drawn from identical probability distributions. We have evaluated the method using synthetic data sets, in which the behavior was modeled as a series of activities, each of which was modeled as a first-order Markov process. By tabulating distributions for pairs of identical and disparate activities, we are able to perform a receiver operating characteristic (ROC) analysis, allowing us to choose appropriate criteria and estimate error rates. Using these criteria, we have applied the methods to data from airline crews, collected in a high-fidelity flight simulator (Gontar & Hoermann, 2014). We conclude by considering the problem of automatic (blind) discovery of activities, using methods developed for text analysis. Anitha Pasupathy University of Washington Visual shape representation in the intermediate stages of the primate brain Decades of research have yielded detailed models of visual form processing in the primary visual cortex (V1) of the primate. Beyond V1, however, our understanding is quite limited. Past studies have shown that in V4, an intermediate stage in the ventral visual pathway, neurons are sensitive to both the curvature of the bounding contour and to the luminance contrast of the stimulus surface relative to the background. These shape selective neurons also display position and size invariance within the confines of the receptive field, but we currently have no model of V4 neurons that can simultaneously achieve all of these stimulus preferences. To attain this elusive goal, we are pursuing two strategies. First, we are conducting paired neurophysiology and modelling experiments aimed at improving the most promising biologically motivated model of V4 form selectivity (the HMAX model), which emphasizes boundary orientation, ignores surface contrast and exhibits limited invariance. For example, we are comparing model units to V4 neurons in terms of whether their shape selectivity is maintained for stimuli defined by an outline alone versus stimuli defined by an outline and surface contrast. To achieve more realistic invariance properties, we are manipulating attributes of the model including normalization equations, design of low-level convolutional features and receptive field density. Second, in parallel to developing this explicitly biologically plausible model, we are also exploring the ability of high-performing artificial object recognition networks to achieve the selectivities and invariances observed in V4. In particular, we have identified and are analyzing the properties of V4-like units in a deep convolutional neural network (AlexNet) trained to recognize objects in NATURAL scenes. We hope to understand what architectures underlie V4-like selectivity, and ultimately what training regimes may be responsible for the emergence of the relevant representations. In my talk, I will present results that reveal how our current best V4 model matches up with V4 physiology, and I will describe plans for future experiments to address the challenges that lie ahead. ================================================================================== Misha Pavel Northeastern University Decomposing Liquid Intelligence We will discuss issues arising when attempting to decompose tasks that are thought to require fluid intelligence(cid:0) into components such as update and inhibit. The notion of these basic components is challenged these executive function(cid:0) tasks are embedded in more real-life-like scenarios such as computer games. We will illustrate these issues using a subset of data from a large study designed to investigate whether fluid intelligence, as measured by tasks such as Ravens Progressive Matrices, can be improved by training the component cognitive functions and their combinations. We plan to discuss both theoretical and practical implications of these results and insights. ================================================================================== 9 Zygmunt Pizlo Purdue University The role of 3D symmetry in figure-ground organization of real scenes. Zygmunt Pizlo, Aaron Michaux and Vijai Jayadevan, Purdue University. The first step in visual perception is determining the presence, the number and the location of objects in front of the observer. In human vision, this step is traditionally called "Figure-Ground Organization (FGO). According to common wisdom, FGO is solved by grouping (clustering) operations on the basis of the similarity of nearby retinal regions such as their similarity with respect to color, motion, texture, etc. The main challenge to this common wisdom derives from the fact that the retinal image always confounds information about the permanent characteristics of the 3D physical objects within a scene with the constantly varying viewing conditions, including illumination, distance, and occlusions. This confound has been an unsurmountable obstacle in formulating a theory of FGO that could produce, even in principle, anything like the level of performance we all achieve in our everyday life. I will describe our recent attempt to develop such a theory. In this theory, the symmetry of objects is the key concept. Symmetry can do the job because (i) all natural objects are characterized by one or more types of 3D symmetry, and (ii) a 3D configuration of unrelated objects is, itself, almost never symmetrical. It follows that detecting 3D symmetries in a real 3D scene will lead to nearly perfect FGO. Nicholas Port Indiana University School of Optometry Ocular-motor Performance of IU Athletes 2.0: A Nefarious Slope Authors: Nicholas Port,Steve Hitzeman, Kacie Monroe, Melissa Elrod-Schmidt Tina It has been suggested that sports ability relates to ocular motor performance. We set out to directly test this hypothesis among ~1400 Indiana University athletes across all 24 sports. Baseline ocular motor data was collected on the first day of training camp just prior to the beginning of the freshman year. Ocular motor tasks included two smooth pursuit tasks (sinusoidal and step-ramp), one self-paced saccade task, and one fixation task (with and without a whole-field motion distractor). Over the ensuing 4 to 7 years of their collegiate athletic careers, we then collected longitudinal data relating to each subject's athletic performance. Seven or more years was sometimes needed in order to obtain a sufficient sample size (e.g., 20 subjects) for sports with small team sizes (e.g., golf and tennis). Large differences were found between sports, with an overall trend for ball sport athletes to have ocular motor performance profiles that differ from non-ball sport athletes e.g. swimming and cross-country). We also found some ocular-motor variables correlate with athletic performance in some sports (e.g. football). Our results, therefore, support the idea that athletic performance is correlated with some smooth pursuit and saccadic ocular motor variables. Additional research is needed to ascertain any causal connections, however. Jenny Read Newcastle University Mantis stereopsis in complex scenes Authors: Vivek Nityananda, Ghaith Tarawneh, Jenny Read Praying mantises are the only non-vertebrate known to possess stereoscopic 3D vision. Yet, very little is known about the capabilities of insect 3D. To date, they have been tested only in very simple scenes containing one or two target objects. Last year at AIC, I presented our lab's technique for displaying 3D stimuli to mantids using blue/green anaglyph glasses. This enables us to probe mantids' response to arbitrarily complex 3D scenes. We have shown that (i) mantids can successfully identify target disparity even in a complex scene containing many moving objects; (ii) mantids can identify disparity of a target object even when that object is perfectly camouflaged on its background in any given frame, and revealed only by its motion. However, we have not yet found evidence that mantids can use disparity to break camouflage when an object is camouflaged even in the motion domain and 10