{"title": "Neurophysiological Evidence of Cooperative Mechanisms for Stereo Computation", "book": "Advances in Neural Information Processing Systems", "page_first": 1201, "page_last": 1208, "abstract": null, "full_text": "Neurophysiological Evidence of Cooperative \n\nMechanisms for Stereo Computation \n\nJason M. Samonds Brian R. Potetz Tai Sing Lee \n\nCenter for the Neural Basis CNBC and Computer CNBC and Computer \n Science Department \n\nof Cognition (CNBC) Science Department \n\nCarnegie Mellon University Carnegie Mellon University Carnegie Mellon University \n\n \n \n \n \n \n \n \n\nPittsburgh, PA 15213 Pittsburgh, PA 15213 Pittsburgh, PA 15213 \n\nsamondjm@cnbc.cmu.edu bpotetz@cs.cmu.edu tai@cnbc.cmu.edu \n\nAbstract \n\nAlthough there has been substantial progress in understanding the neuro-\nphysiological mechanisms of stereopsis, how neurons interact in a network \nduring stereo computation remains unclear. Computational models on \nstereopsis suggest local competition and long-range cooperation are impor-\ntant for resolving ambiguity during stereo matching. To test these predic-\ntions, we simultaneously recorded from multiple neurons in V1 of awake, \nbehaving macaques while presenting surfaces of different depths rendered \nin dynamic random dot stereograms. We found that the interaction between \npairs of neurons was a function of similarity in receptive fields, as well as \nof the input stimulus. Neurons coding the same depth experienced common \ninhibition early in their responses for stimuli presented at their non-\npreferred disparities. They experienced mutual facilitation later in their re-\nsponses for stimulation at their preferred disparity. These findings are con-\nsistent with a local competition mechanism that first removes gross mis-\nmatches, and a global cooperative mechanism that further refines depth es-\ntimates. \n\n1 Introduction \nThe human visual system is able to extract three-dimensional (3D) structures in random \nnoise stereograms even when such images evoke no perceptible patterns when viewed \nmonocularly [1]. Bela Julesz proposed that this is accomplished by a stereopsis mechanism \nthat detects correlated shifts in 2D noise patterns between the two eyes. He also suggested \nthat this mechanism likely involves cooperative neural processing early in the visual system. \nMarr and Poggio formalized the computational constraints for solving stereo matching (Fig. \n1a) and devised an algorithm that can discover the underlying 3D structures in a variety of \nrandom dot stereogram patterns [2]. Their algorithm was based on two rules: (1) each ele-\nment or feature is unique (i.e., can be assigned only one disparity) and (2) surfaces of objects \nare cohesive (i.e., depth changes gradually across space). To describe their algorithm in neu-\nrophysiological terms, we can consider neurons in primary visual cortex as simple element \nor feature detectors. The first rule is implemented by introducing competitive interactions \n(mutual inhibition) among neurons of different disparity tuning at each location (Fig. 1b, \nblue solid horizontal or vertical lines), allowing only one disparity to be detected at each \nlocation. The second rule is implemented by introducing cooperative interactions (mutual \nfacilitation) among neurons tuned to the same depth (image disparity) across different spatial \nlocations (Fig. 1b, along the red dashed diagonal lines). In other words, a disparity estimate \nat one location is more likely to be correct if neighboring locations have similar disparity \nestimates. A dynamic system under such constraints can relax to a stable global disparity \nmap. Here, we present neurophysiological evidence of interactions between disparity-tuned \n\n\fneurons in the primary visual cortex that is consistent with this general approach. We sam-\npled from a variety of spatially distributed disparity tuned neurons (see electrodes Fig. 1b) \nwhile displaying DRDS stimuli defined at various disparities (see stimulus Fig.1b). We then \nmeasured the dynamics of interactions by assessing the temporal evolution of correlation in \nneural responses. \n\n Left Image \n\n Right Image \n\na \n? \n\nb\n\n \n\ne\ng\na\nm\n\nI\n \nt\nf\ne\nL\n\nElectrodes \n\nDisparity \n\nStimulus \n\nRight Image \n\nFigure 1: (a) Left and right images of random dot stereogram (right image has been shifted \nto the right). (b) 1D graphical depiction of competition (blue solid lines) and cooperation \n(red dashed lines) among disparity-tuned neurons with respect to space as defined by Marr \nand Poggio\u2019s stereo algorithm [2]. \n\n2 Methods \n\n2.1 Recording and stimulation \nRecordings were made in V1 of two awake, behaving macaques. We simultaneously re-\ncorded from 4-8 electrodes providing data from up to 10 neurons in a single recording ses-\nsion (some electrodes recorded from as many as 3 neurons). We collected data from 112 \nneurons that provided 224 pairs for cross-correlation analysis. For stimuli, we used 12 Hz \ndynamic random dot stereograms (DRDS; 25% density black and white pixels on a mean \nluminance background) presented in a 3.5-degree aperture. Liquid crystal shutter goggles \nwere used to present random dot patterns to each eye separately. Eleven horizontal dispari-\nties between the two eyes, ranging from \u00b10.9 degrees, were tested. Seventy-four neurons \n(66%) had significant disparity tuning and 99 pairs (44%) were comprised of neurons that \nboth had significant disparity tuning (1-way ANOVA, p<0.05). \n \n\na \n\nb \n\n \nr\no\n\ni\nr\ne\nt\nn\nA\n\n \n-\n \nr\no\n\ni\nr\ne\nt\ns\no\nP\n\n5mm \nMedial - Lateral \n\n100\u00b5V \n0.2ms\n\n1 \u00b01 \u00b0\n\nFigure 2: (a) Example recording session from five electrodes in V1. (b) Receptive field \n(white box\u2014arrow represents direction preference) and random dot stereogram locations for \nsame recording session (small red square is the fixation spot). \n\n\f2.2 Data analysis \nInteraction between neurons was described as \"effective connectivity\" defined by cross-\ncorrelation methods [3]. First, the probability of all joint spikes (x and y) between the two \nneurons was calculated for all times from stimulus onset (t1 and t2) including all possible lag \ntimes (t1 - t2) between the two neurons (2D joint peristimulus time histogram\u2014JPSTH). \nNext, the cross-product of each neuron\u2019s PSTH (joint probabilities expected from chance) \nwas subtracted from the JPSTH; this difference is referred to as the cross-covariance histo-\ngram. Finally, the cross-covariance histogram was normalized by the geometric mean of the \nauto-covariance histograms: \n\n \n\n \n\n)t,t(C\n2\n\ny,x\n\n1\n\n=\n\n(\n\n)t(x)t(x\n1\n\n1\n\n\u2212\n\n1\n\n)t(y)t(x\n\u2212\n2\n)(\n)t(x)t(x\n1\n\n1\n\n1\n\n)t(y)t(x\n2\n)t(y)t(y\n2\n\n2\n\n\u2212\n\n))t(y)t(y\n\n2\n\n2\n\n \n\n (1) \n\nThis normalized cross-covariance histogram is a 2D matrix of Pearson\u2019s correlation coeffi-\ncients between the two neurons where the axes represent time from stimulus onset (Figure \n3). The principal diagonal also represents time from stimulus onset for correlation and the \nopposite diagonal represents lag time between the two neurons. We derived three measure-\nments from this matrix to describe the \u201ceffective connectivity\u201d between neuron pairs. Using \nbootstrapped samples of stimulus trials, we estimated 95% confidence intervals for these \nthree measurements [4]. We first integrated along the principal diagonal to produce correla-\ntion versus lag time (i.e., the traditional cross-correlation histogram\u2014CCH). We used CCHs \nto find significant correlation at or near 0 ms lag times (suggesting synaptic connectivity \nbetween the neurons). Second, we integrated under the half-height full bandwidth of signifi-\ncant correlation peaks to quantify effective connectivity. Figure 4 shows the population av-\nerage of normalized CCHs (n = 27) and 95% confidence intervals. Finally, we repeated this \nintegration along the principal diagonal to obtain the temporal evolution of effective connec-\ntivity (computed with a running 100 ms window). \n\nCCH \n\nJPSTH \n\n \n2\nt\n\n \n)\n2\nt\n(\ny\n\nx(t1) \n\nt1 \n\nFigure 3: Example normalized cross-covariance histogram. \n\nIn computing effective connectivity with Equation 1, we assume trial-to-trial stationarity. If \nthis is not true (e.g., due to difference in attentional effort in different trials), correlation \npeaks can emerge that are not due to effective connectivity [5]. We applied a correction to \nequation 1 [5,6] based on the average firing rate for each trial. However, no significant dif-\nference in correlation peaks was observed. In addition, changes in DRDS properties other \nthan disparity did not cause significant changes to correlation peak properties. Finally, al-\nternative cross-correlation methods (CCG) [7] using responses to the same exact random dot \npattern to predict correlation expected from chance, again, lead to no significant difference \nin correlation peak properties. These observations justify our assumption that the effective \nconnectivity computed in our case does not arise due to trial-to-trial non-stationarity. \n\n \n\n\f0.008 \n\na \n\n \n \n\nn\no\n\ni\nt\na\n\nl\n\ne\nr\nr\no\nC\n\n0.004 \n\n0.000 \n\n-0.004 \n\n-300 \n\n-100 \n100\nLag Time (ms)\n\n300\n\n b\n0.006\n0.004\n0.002\n0.000\n-0.002\n\n-50\n\n-25\n\n0.17 \u00b1 0.02\n\n0\n\n25 \n\n50 \n\nHalf-Height \nHalf Bandwidth \nPeak Lag Time \n\n0 \n\n25 \nLag Time (ms) \n\n50 \n\nFigure 4: (a) Population average CCH for 27 neuron pairs with a significant correlation \npeak. (b) Same as (a), but zoomed into \u00b150 ms lag times with statistics of peak properties \n(mean \u00b1 s.e.m.). \n\n3 Interaction depends on tuning properties \nThe primary indicator of whether or not a neuron pair had a significant correlation peak at or \nnear a 0 ms lag time, for this class of stimuli, was similarity in disparity tuning between the \ntwo neurons. Neuron pairs with significant correlation peaks (n = 27; 27%) tended to have \nmore similar disparity peaks, bandwidths, and frequencies (determined from fitted Gabor \nfunctions) than neuron pairs that did not have significant correlation peaks. We quantified \nsimilarity in tuning using the similarity index (SI), which is Pearson\u2019s product-moment cor-\nrelation [8]: \n \n\ny)(x\n\n\u2211\n\n)y\n\nx(\n\nSI\n\n=\n\n\u2212\n\n\u2212\n\n1i\n=\n\nn\n\ni\n\ni\n\n\u239b\n\u239c\n\u239c\n\u239d\n\nn\n\n\u2211\n\nx(\n\n\u2212\n\n2\n\n)x\n\n\u239e\n\u239b\n\u239f\n\u239c\n\u239f\n\u239c\n\u23a0\n\u239d\n\nn\n\n\u2211\n\ny(\n\n\u2212\n\n2\n\n)y\n\n\u239e\n\u239f\n \n\u239f\n\u23a0\n\ni\n\ni\n\n \n\n \n\n1i\n=\n\n1i\n=\n\n (2) \nwhere i is each point on the disparity tuning curve, x and y are the firing rates at each point \nfor each neuron, and x and y are the mean firing rates across the tuning curve. \nFigure 5a and 5b clearly show that both the probability of correlation and strength in correla-\ntion increase with greater SI (n = 27 pairs). This relationship is limited to long-range inter-\nactions among neurons because our electrodes were all at least 1 mm apart. This suggests \nthey are likely mediated by the well known long-range intracortical connections in V1 that \nlink neurons of similar orientation across space [9]. Our results suggest that these connec-\ntions might also be shared to link similar disparity neurons together. Because connectivity \nalso depended on orientation (Figure 5c), V1 connectivity among neurons appears to depend \non similarity across multiple cue dimensions. \n\n0.4 \n0.3 \n0.2 \n0.1 \n0 \n\n \n)\n7\n2\n=\nn\n(\n \n\ns\nr\ni\n\na\nP\n\n \n\ne\ng\na\nt\nn\ne\nc\nr\ne\nP\n\na \n\n-0.9 \nSimilarity Index (SI) \n\n-0.3 \n\n0.3 \n\nb\n\n \n\nn\no\n\ni\nt\na\nl\ne\nr\nr\no\nC\n\n0.9 \n\n-1.0\n\n-0.5\n\n0.4\n\n0.3\n\n0.2\n\n0.1\n\n0.0\n\nr = 0.49 \np = 0.01 \n\nc\n\nr = -0.40 \np = 0.04 \n\n0.0\nSI\n\n0.5\n\n1.0\n\n0\n\n20\n\n40 \n\n60 \n\n80\n\n \u2206 Orientation Preference \n\nFigure 5: (a) Likelihood of significant correlation peak with respect to similar disparity tun-\ning. (b) Strength of correlation increases with similarity. (c) Correlation is also more likely \nif orientation preference is similar. \n\n\fFrom the 12 pairs of neurons recorded on a single electrode, correlation was observed among \nneuron pairs with very similar disparity tuning as well as among neurons with nearly oppo-\nsite disparity tuning (see also [8]). This suggests that antagonistic disparity-tuned neurons \ntend to spatially coexist, and their interactions are likely competitive. \n\n4 Interaction is stimulus-dependent \nThe interaction between pairs of neurons was not simply a function of the similarity between \ntheir receptive field properties but was also a function of the input stimuli (or stimulus dis-\nparity in our case). The effective connectivity was significantly modulated (1-way ANOVA, \np<0.05) by the stimulus disparity for 25 out of the 27 pairs. We are not suggesting synaptic \nconnections physically change, but rather that the effectiveness of those connections can \nchange depending on the spiking activity and therefore the stimulus input. For neuron pairs \nwith similar disparity tuning, the strongest correlation was observed at their shared preferred \ndisparity, i.e. the peak of the disparity tuning curves based on firing rate (as shown in Figure \n6). This suggests facilitation is strongest when a frontal parallel plane activated these neu-\nrons simultaneously at their preferred depth. As the stimulus plane moved away from this \ndepth, the effective connectivity between the neurons became weaker. This was observed in \n10 pairs (e.g., Figure 6c). For the other 17 pairs (e.g., Figure 6d), the correlation or effective \nconnectivity was again strongest at the neuron pair's shared preferred disparity. However, \nthese pairs in addition exhibited secondary correlation peaks for disparity stimuli that pro-\nduced the lowest firing rates (even below the baseline for DRDSs). \n\na \n\n50 sps \n\n80 sps \n\nb \n\n20 sps \n\n40 sps \n\n \n\ne\nt\na\nR\ng\nn\n\n \n\ni\nr\ni\n\nF\n\n0.0\n\n0.9\n\n-0.9\n\n0.15\n\nd \n\n0.0\n\n0.9 \n\n \n \n \n \n \n \n \n \n\n0.30 \n\n-0.9 \nc \n\n0.20 \n\n0.10 \n\n \n\nn\no\n\ni\nt\na\n\nl\n\ne\nr\nr\no\nC\n\n \n \n \n \n \n \n \n \n \n \n \n \n\n0.10\n\n0.05\n\n0.00\n\n-0.9\n\n0.9 \nHorizontal Disparity (\u00b0) \n\n0.0\n\n0.00 \n\n-0.9 \n\n0.9\nHorizontal Disparity (\u00b0)\n\n0.0\n\nFigure 6: Top row are disparity tuning curves based on firing rates (mean \u00b1 s.e.m.). Bottom \nrow are disparity tuning curves based on correlation for the corresponding pairs of neurons \nin the top row. Error bars are 95% confidence intervals and dashed lines represent 95% con-\nfidence of the mean correlation. \nCross-correlation peaks are interpreted as a result of effective circuits that may represent any \ncombination of a variety of synaptic connections that may have a bias in direction (one neu-\nron drives the other) or may not have a bias in direction (zero lag time; both neurons receive \na common drive) [10]. As correlation peaks become broader, as in our case (mean = 42 ms), \nthis interpretation becomes more ambiguous (more possible circuits). The broader positive \ncorrelation peaks can even be caused by common inhibitory circuitry. One way to poten-\ntially disambiguate our interpretations is to consider firing rate behavior. The positive corre-\nlation measured at the preferred disparity suggests that the interaction was likely facilitatory \nin nature based on the increased firing of the neurons. The positive correlation measured at \nthe disparity where both neurons' firing rates were depressed, i.e. at the valley of the firing-\n\n\frate based disparity tuning curves, suggests that the correlation likely arose from common \ninhibition (presumably from neurons that preferred that disparity). \n\n5 Temporal dynamics of interaction \nWe can compare the temporal dynamics of the correlation with the temporal dynamics of the \nfiring rate of the neurons to gain more insight into the possible underlying circuitry. We \ncomputed the correlation every 1 msec over a 100 ms running window, and found that the \ncorrelation peak at the preferred disparity (based on firing rates) occurred at a later time \n(250-350 ms post-stimulus onset) than the correlation peaks at the non-preferred disparity \n(100-200 ms). Figure 7 illustrates the temporal dynamics of correlation for the example neu-\nron pair shown in Figure 6b and 6d. The distinct interval in which correlation emerged at \nthe preferred and the non-preferred disparities was consistently observed for all 27 pairs of \nneurons. Even for the example shown in Figure 6c, there were peaks in correlation in the \nearly part of the response at the most non-preferred disparities. The timing of these two \nphases of correlation was also rather consistent over the population of pairs. \n\n400 \n\n300 \n\n200 \n\n100 \n\n \n)\ns\nm\n\n(\n \n\ne\nm\nT\n\ni\n\n \n\nn\no\n\ni\nt\na\nl\ne\nr\nr\no\nC\n\n0.3\n\n0.2\n\n0.1\n\n0\n\n-0.1\n0.3\n\n0.2\n\n0.1\n\n \n\nn\no\n\ni\nt\na\n\nl\n\ne\nr\nr\no\nC\n\n0.4 \n\n0.2 \n\n0.0\nCorrelation \n\n-0.9 \n\n0\n\nHorizontal Disparity (\u00b0) \n\n0.9 \n\n0\n-0.9\n\n-0.1\n\n0 \n\n0.9 \nHorizontal Disparity (\u00b0) \n\nFigure 7: Temporal dynamics of correlation for example neuron pair shown in Figure 6, \nright. From left to right: Correlation versus time for preferred (red) and non-preferred (blue) \ndisparities. Contour map of correlation versus time and disparity. Disparity tuning based on \ncorrelation for the early (blue) and late (red) portion of the response (95% confidence inter-\nvals). Correlation was calculated every 1 ms over 100 ms windows. \nBy examining the interplay between firing rate and correlation, we were able to gain even \ngreater insight about the interactions among neuron pairs. To summarize this interplay across \nour population, we compared the temporal evolution of the correlation at three distinct dis-\nparities with the temporal evolution of the firing rates at the same disparities (also smoothed \nwith 100 ms time windows). The first disparity, the preferred disparity A, is where we \nmeasured the strongest correlation and was at or near the highest firing rate measured in in-\ndividual neurons (see Figure 8, left). The second important disparity, the most non-preferred \ndisparity C, was where we measured secondary correlation peaks and coincided with the \nlowest firing rates observed in individual neurons. Lastly, we looked at a disparity B that \nwas in between disparities A and C. \nFigure 8 shows that neurons responded better to their preferred disparity over other dispari-\nties very early, resulting in immediate moderate firing rate-based disparity tuning. Then \nshortly after (100 ms), a correlation peak emerges at the least preferred disparity C. This \ncoincides with suppression of firing rate for all disparities (Figure 8, blue dashed line). \nHowever, the suppression in firing rate is much stronger for C where the firing rate diverges \ndownward from the firing rates for A and B sharpening the disparity tuning (Figure 8, blue \narrow; see also [11]). The strong correlation coupled with the decrease in firing suggests \nstrong common inhibition. \n\n\fCorrelation \n\nA \n\nB \n\nC \n\nFiring Rate \n\nB \n\nA \n\nC \n\n-100\n\n \n)\ns\np\ns\n(\n \n\ne\nt\na\nR\ng\nn\n\n \n\ni\nr\ni\n\nF\n\n-100\n\n0\n\n-0.05\n30\n\n25\n20\n15\n10\n5\n0\n\nn = 27 pairs \n\nA\nB\nC\n\n \n\nn\no\n\ni\nt\na\n\nl\n\ne\nr\nr\no\nC\n\n0.25\n\n0.20\n\n0.15\n\n0.10\n\n0.05\n\n0.00\n\n100\n\n200\n\n300 \n\n400 \n\nn = 32 cells \n\nA\nB\nC\n\n0\n\n100\n\n200\n\nTime (ms)\n\n300 \n\n400 \n\nFigure 8: Population average of normalized correlation versus time (top) for three disparities \nshown on the left. Population average of normalized PSTHs for same three disparities (bot-\ntom). Both correlation and firing rates were calculated every 1 ms over 100 ms windows. \nOnce the correlation peak at C subsided (200 ms), the correlation increased for A (red \ndashed line). When the correlation for A peaked, the correlation decreased for B and C, lead-\ning to very sharp correlation-based disparity tuning (see also Figure 7). This correlation-\nbased tuning can facilitate depth estimates by changing how effectively these signals are \nintegrated downstream as a function of disparity [12]. \nOur interpretation is that the initial firing rate bias leads to antagonistic disparity-tuned neu-\nrons generating common inhibition that suppresses firing at non-preferred disparities, re-\nmoving potential mismatches. The immediacy suggests that mutual inhibition was local, \nwhich is consistent with our observation that many opposing disparity-tuned neurons spa-\ntially coexisted. The slower correlation peak at the preferred disparity A is indicative of mu-\ntual facilitation that occurred when the depth estimates of spatially distinct neurons matched. \nThis facilitation leads to a more precise estimate of depth. \n\n6 Discussion and conclusions \nThe findings from this study provide support to Julesz\u2019s proposal that cooperative and com-\npetitive mechanisms in primary visual cortex are utilized for estimating global depth in ran-\ndom dot stereograms [1], which was later described formally by Marr and Poggio [2]. More \nrecent cooperative stereo computation models allow excitatory interaction between neurons \nof different disparities separated by long distance. This is used to accommodate the compu-\ntation of slanted surfaces [13,14]. In this experiment, we only tested frontal parallel planes, \nthus, we cannot answer whether or not effective connections and facilitation exist between \nneurons with larger disparity differences over long distances. This will require further ex-\nperiments using planes with disparity gradients. \nThe observation that initial correlation peaks occurred at disparities that evoked the lowest \nfiring rates in neurons, suggests that correlation peaks emerged from common inhibition for \nnon-preferred disparities. The observation that later correlation occurred at disparities that \nevoked the highest firing rates suggests that neurons were mutually exciting each other at \ntheir preferred disparity. Our neurophysiological data reveal interesting dynamics between \nnetwork-based (effective connectivity) and firing rate-based encoding of depth estimates. \nThe observation that inhibition precedes facilitation suggests that competition is local (re-\n\n\fcalling neurons at the same electrode tend to have opposite disparity tuning) and cooperation \nis more global (mediated through long-range connectivity). Local competition between neu-\nrons encoding different depths is consistent with the uniqueness principle of Marr and Pog-\ngio's algorithm [2]. In addition, cooperation among neurons encoding the same depth across \nspace was predicted by the second rule of their algorithm: matter is cohesive. These two \ninteractions are robust at removing potential ambiguity during stereo matching and depth \ninference. \nPrevious neurophysiological data had suggested that intracortical connectivity in primary \nvisual cortex underlies competitive [15] and cooperative [16] mechanisms for improving \nestimates of orientation. Our data suggests similar circuitry might play a role also in stereo \nmatching [17]. However, this study is distinct in that it provides detailed empirical support \nfor computational algorithms for solving stereo matching. It thus highlights the importance \nof computational algorithms in generating hypotheses to guide future neurophysiological \nstudies. \n\nAcknowledgments \nWe thank George Gerstein and Jeff Keating for JPSTH software. Supported by NIMH IBSC \nMH64445 and NSF CISE IIS-0413211 grants. \n\n Science \n\nReferences \n[1] Julesz, B. (1971) Foundations of cyclopean perception. Chicago: University of Chicago Press. \n[2] Marr, D. & Poggio, T. (1976) Cooperative computation of stereo disparity. \n194(4262):283-287. \n[3] Aertsen, A.M., Gerstein, G.L., Habib, M.K. & Palm, G. (1989) Dynamics of neuronal firing corre-\nlation: modulation of \"effective connectivity\". Journal of Neurophysiology 61(5):900-917. \n[4] Efron, B. & Tibshirani, R. (1993) An Introduction to the Bootstrap. New York: Chapman & Hall. \n[5] Brody, C.D. (1999) Correlations without synchrony. Neural Computation 11(7):1537-1551. \n[6] Gerstein, G.L. & Kirkland, K.L. (2001) Neural assemblies: technical issues, analysis, and \nmodeling. Neural Networks 14(6-7):589-598. \n[7] Kohn, A. & Smith, M.A. (2005) Stimulus dependence of neuronal correlation in primary visual \ncortex of the macaque. Journal of Neuroscience 25(14):3661-3673. \n[8] Menz, M. & Freeman, R.D. (2004) Functional connectivity of disparity-tuned neurons in the visual \ncortex. Journal of Neurophysiology 91(4):1794-1807. \n[9] Gilbert, C.D. & Wiesel, T.N. (1989) Columnar specificity of intrinsic horizontal and corticocorti-\ncal connections in cat visual cortex. Journal of Neuroscience 9(7):2432-2442. \n[10] Moore, G.P., Segundo, J.P., Perkel, D.H. & Levitan, H. (1970) Statistical signs of synaptic inter-\naction in neurons. Biophysics Journal 10(9):876-900. \n[11] Menz, M. & Freeman, R.D. (2003) Stereoscopic depth processing in the visual cortex: a coarse-\nto-fine mechanism. Nature Neuroscience 6(1):59-65. \n[12] Bruno, R.M. & Sakmann, B. (2006) Cortex is driven by weak but synchronously active thalamo-\ncortical synapses. Science 312(5780):1622-1627. \n[13] Prazdny, K. (1985) Detection of binocular disparities. Biological Cybernetics 52(2):93-99. \n[14] Pollard, S.B., Mayhew, J.E., & Frisby, J.P. (1985) PMF: a stereo correspondence algorithm \nusing a disparity gradient limit. Perception 14(4):449-470. \n[15] Ringach, D.L., Hawken, M.J. & Shapley, R. (1997) Dynamics of orientation tuning in macaque \nprimary visual cortex. Nature 387(6630):281-284. \n[16] Samonds, J.M., Allison, J.D., Brown, H.A. & Bonds, A.B. (2004) Cooperative synchronized as-\nsemblies enhance orientation discrimination. Proceedings of the National Academy of Sciences USA \n101(17):6722-6727. \n[17] Ben-Shahar, O., Huggins, P.S., Izo, T. & Zucker, S.W. (2003) Cortical connections and early \nvisual function: intra- and inter-columnar processing. Journal of Physiology (Paris) 97(2-3):191-208. \n\n\f", "award": [], "sourceid": 3011, "authors": [{"given_name": "Jason", "family_name": "Samonds", "institution": null}, {"given_name": "Brian", "family_name": "Potetz", "institution": null}, {"given_name": "Tai", "family_name": "Lee", "institution": null}]}