Transformations of sensory information in the brain suggest changing criteria for optimality.

Neurons throughout the brain modulate their firing rate lawfully in response to sensory input. Theories of neural computation posit that these modulations reflect the outcome of a constrained optimization in which neurons aim to robustly and efficiently represent sensory information. Our understanding of how this optimization varies across different areas in the brain, however, is still in its infancy. Here, we show that neural sensory responses transform along the dorsal stream of the visual system in a manner consistent with a transition from optimizing for information preservation towards optimizing for perceptual discrimination. Focusing on the representation of binocular disparities-the slight differences in the retinal images of the two eyes-we re-analyze measurements characterizing neuronal tuning curves in brain areas V1, V2, and MT (middle temporal) in the macaque monkey. We compare these to measurements of the statistics of binocular disparity typically encountered during natural behaviors using a Fisher Information framework. The differences in tuning curve characteristics across areas are consistent with a shift in optimization goals: V1 and V2 population-level responses are more consistent with maximizing the information encoded about naturally occurring binocular disparities, while MT responses shift towards maximizing the ability to support disparity discrimination. We find that a change towards tuning curves preferring larger disparities is a key driver of this shift. These results provide new insight into previously-identified differences between disparity-selective areas of cortex and suggest these differences play an important role in supporting visually-guided behavior. Our findings emphasize the need to consider not just information preservation and neural resources, but also relevance to behavior, when assessing the optimality of neural codes.

Transformations of sensory information in the brain reflect a changing definition of optimality.

Neurons throughout the brain modulate their firing rate lawfully in response to changes in sensory input. Theories of neural computation posit that these modulations reflect the outcome of a constrained optimization: neurons aim to efficiently and robustly represent sensory information under resource limitations. Our understanding of how this optimization varies across the brain, however, is still in its infancy. Here, we show that neural responses transform along the dorsal stream of the visual system in a manner consistent with a transition from optimizing for information preservation to optimizing for perceptual discrimination. Focusing on binocular disparity - the slight differences in how objects project to the two eyes - we re-analyze measurements from neurons characterizing tuning curves in macaque monkey brain regions V1, V2, and MT, and compare these to measurements of the natural visual statistics of binocular disparity. The changes in tuning curve characteristics are computationally consistent with a shift in optimization goals from maximizing the information encoded about naturally occurring binocular disparities to maximizing the ability to support fine disparity discrimination. We find that a change towards tuning curves preferring larger disparities is a key driver of this shift. These results provide new insight into previously-identified differences between disparity-selective regions of cortex and suggest these differences play an important role in supporting visually-guided behavior. Our findings support a key re-framing of optimal coding in regions of the brain that contain sensory information, emphasizing the need to consider not just information preservation and neural resources, but also relevance to behavior.

Ocular-following responses in school-age children.

Ocular following eye movements have provided insights into how the visual system of humans and monkeys processes motion. Recently, it has been shown that they also reliably reveal stereoanomalies, and, thus, might have clinical applications. Their translation from research to clinical setting has however been hindered by their small size, which makes them difficult to record, and by a lack of data about their properties in sizable populations. Notably, they have so far only been recorded in adults. We recorded ocular following responses (OFRs)-defined as the change in eye position in the 80-160 ms time window following the motion onset of a large textured stimulus-in 14 school-age children (6 to 13 years old, 9 males and 5 females), under recording conditions that closely mimic a clinical setting. The OFRs were acquired non-invasively by a custom developed high-resolution video-oculography system, described in this study. With the developed system we were able to non-invasively detect OFRs in all children in short recording sessions. Across subjects, we observed a large variability in the magnitude of the movements (by a factor of 4); OFR magnitude was however not correlated with age. A power analysis indicates that even considerably smaller movements could be detected. We conclude that the ocular following system is well developed by age six, and OFRs can be recorded non-invasively in young children in a clinical setting.

Weighted summation and contrast normalization account for short-latency disparity vergence responses to white noise stimuli in humans.

Natural images are typically broadband, whereas detectors in early visual processing are selective for narrow ranges of spatial frequency. White noise patterns are widely used in laboratory settings to investigate how responses are derived from Fourier components in the image. Here, we report disparity vergence responses (DVRs) to white noise stimuli in human subjects and compare these with responses to white noise patterns filtered with bandpass filters and notch filters and to sinusoidal gratings. Although the contribution of these short-latency eye movements to the overall vergence response to a given stimulus is generally small, they have proven to be a valuable tool for the study of the early mechanisms that process disparity stimuli in human subjects. Removing lower spatial frequency (SF) components reduced DVR amplitude, whereas removing higher SF components led to an increase in DVR amplitude. For larger disparities, the transition occurred at lower SFs. All of these effects were quantitatively well described by a model that combined two factors: (a) an excitatory drive determined by a weighted sum of stimulus Fourier components, which was scaled by (b) a contrast normalization mechanism.

Pattern Motion Direction Is Encoded in the Population Activity of Macaque Area MT.

Direction selective neurons in macaque primary visual cortex are narrowly tuned for orientation, and are thus afflicted by the aperture problem. At the next stage of motion processing, in the middle temporal (MT) area, some cells appear to solve this problem, responding to the pattern motion direction of plaids. Models have been proposed to account for this computation, but they do not replicate the diversity of responses observed in MT. We recorded from 386 cells in area MT of two male macaques, while presenting a wide range of random-line stimuli and their compositions into noise plaids. As we broadened the range of stimuli used to probe the cells, yielding ever more challenging conditions for extracting pattern motion, the diversity of the responses observed increased, and the fraction of cells that faithfully encoded pattern motion direction shrank. However, we show here that a pattern motion signal is present at the population level. We identified four mechanisms, one never proposed before, that together might account for the observed diversity in single-cell responses. Pattern motion is thus extracted in area MT, but it is encoded across the population, and not in a small subset of pattern neurons.SIGNIFICANCE STATEMENT Some neurons in the middle temporal area of macaques solve the aperture problem, signaling the direction of motion of complex patterns. As the number of pattern types used to probe this mechanism is increased, fewer and fewer cells retain this capability. We show here that different cells fail in different ways, and that simply summing their responses averages away their failures, yielding a clear pattern motion signal. Similar encodings, which unequivocally violate the "neuron as a feature detector" hypothesis that has dominated sensory processing theories for the past 50 years, might apply throughout the brain.

Model-based characterization of the selectivity of neurons in primary visual cortex.

Statistical models are increasingly being used to understand the complexity of stimulus selectivity in primary visual cortex (V1) in the context of complex time-varying stimuli, replacing averaging responses to simple parametric stimuli. Although such models often can more accurately reflect the computations performed by V1 neurons in more natural visual environments, they do not by themselves provide insight into V1 neural selectivity to basic stimulus features such as receptive field size, spatial frequency tuning, and phase invariance. Here, we present a battery of analyses that can be directly applied to encoding models to link complex encoding models to more interpretable aspects of stimulus selectivity. We apply this battery to nonlinear models of V1 neurons recorded in awake macaque during random bar stimuli. In linking model properties to more classical measurements, we demonstrate several novel aspects of V1 selectivity not available to simpler experimental measurements. For example, this approach reveals that individual spatiotemporal elements of the V1 models often have a smaller spatial scale than the neuron as a whole, resulting in nontrivial tuning to spatial frequencies. In addition, we propose measures of nonlinear integration that suggest that classical classifications of V1 neurons into simple versus complex cells will be spatial-frequency dependent. In total, rather than obfuscate classical characterizations of V1 neurons, model-based characterizations offer a means to more fully understand their selectivity, and link their classical tuning properties to their roles in more complex, natural, visual processing.NEW & NOTEWORTHY Visual neurons are increasingly being studied with more complex, natural visual stimuli, and increasingly complex models are necessary to characterize their response properties. Here, we describe a battery of analyses that relate these more complex models to classical characterizations. Using such model-based characterizations of V1 neurons furthermore yields several new insights into V1 processing not possible to capture in more classical means to measure their visual selectivity.

Short-latency ocular following responses to motion stimuli are strongly affected by temporal modulations of the visual content during the initial fixation period.

Neuronal and psychophysical responses to a visual stimulus are known to depend on the preceding history of visual stimulation, but the effect of stimulation history on reflexive eye movements has received less attention. Here, we quantify these effects using short-latency ocular following responses (OFRs), a valuable tool for studying early motion processing. We recorded, in human subjects, the horizontal OFRs induced by drifting vertical 1D pink noise. The stimulus was preceded by 600 to 1000 ms of maintained fixation (on a visible cross), and we explored the effect of different stimuli ("fixation patterns") presented during the fixation period. We found that any temporal modulation present during the fixation period reduced the magnitude of the subsequent OFRs. Even changes in the overall luminance during the fixation period induced significant suppression. The magnitude of the effect was a function of both spatial and temporal structure of the fixation pattern. Suppression that was selective for both relative orientation and relative spatial frequency accounted for a considerable fraction of total suppression. Finally, changes in stimulus temporal structure alone (i.e. "flicker" versus "transparent motion") led to changes in the spatial frequency tuning of suppression. In the time domain, the suppression developed quickly: 100 ms of temporal modulation in the fixation pattern produced up to 80% of maximal suppression. Recovery from suppression was instead more gradual, taking up to several seconds. By presenting transparent motion during the fixation period, with opposite motion signals having different spatial frequency content, we also discovered a direction-selective component of suppression, which depended on both the frequency and the direction of the moving stimulus.

Short-latency ocular-following responses: Weighted nonlinear summation predicts the outcome of a competition between two sine wave gratings moving in opposite directions.

We recorded horizontal ocular-following responses to pairs of superimposed vertical sine wave gratings moving in opposite directions in human subjects. This configuration elicits a nonlinear interaction: when the relative contrast of the gratings is changed, the response transitions abruptly between the responses elicited by either grating alone. We explore this interaction in pairs of gratings that differ in spatial and temporal frequency and show that all cases can be described as a weighted sum of the responses to each grating presented alone, where the weights are a nonlinear function of stimulus contrast: a nonlinear weighed summation model. The weights depended on the spatial and temporal frequency of the component grating. In many cases the dominant component was not the one that produced the strongest response when presented alone, implying that the neuronal circuits assigning weights precede the stages at which motor responses to visual motion are generated. When the stimulus area was reduced, the relationship between spatial frequency and weight shifted to higher frequencies. This finding may reflect a contribution from surround suppression. The nonlinear interaction is strongest when the two components have similar spatial frequencies, suggesting that the nonlinearity may reflect interactions within single spatial frequency channels. This framework can be extended to stimuli composed of more than two components: our model was able to predict the responses to stimuli composed of three gratings. That this relatively simple model successfully captures the ocular-following responses over a wide range of spatial/temporal frequency and contrast parameters suggests that these interactions reflect a simple mechanism.

Motion and binocular disparity processing: Two sides of two different coins.

From a mathematical point of view, extracting motion and disparity signals from a binocular visual stream requires very similar operations, applied over time for motion and across eyes for disparity. This similarity is reflected in the theories that have been proposed to describe the neural mechanisms used by the brain to extract these signals. At the behavioral level there are, however, several differences in how humans react to these stimuli, which presumably reflect differences in how these signals are processed by the brain. Here we highlight three such differences: the degree to which different axes of motion/disparity are treated isotropically, the importance of reference signals, and the rules that underlie the combination of 1D signals to extract 2D signals.

The psychophysics of stereopsis can be explained without invoking independent ON and OFF channels.

Early vision proceeds through distinct ON and OFF channels, which encode luminance increments and decrements respectively. It has been argued that these channels also contribute separately to stereoscopic vision. This is based on the fact that observers perform better on a noisy disparity discrimination task when the stimulus is a random-dot pattern consisting of equal numbers of black and white dots (a "mixed-polarity stimulus," argued to activate both ON and OFF stereo channels), than when it consists of all-white or all-black dots ("same-polarity," argued to activate only one). However, it is not clear how this theory can be reconciled with our current understanding of disparity encoding. Recently, a binocular convolutional neural network was able to replicate the mixed-polarity advantage shown by human observers, even though it was based on linear filters and contained no mechanisms which would respond separately to black or white dots. Here, we show that a subtle feature of the way the stimuli were constructed in all these experiments can explain the results. The interocular correlation between left and right images is actually lower for the same-polarity stimuli than for mixed-polarity stimuli with the same amount of disparity noise applied to the dots. Because our current theories suggest stereopsis is based on a correlation-like computation in primary visual cortex, this postulate can explain why performance was better for the mixed-polarity stimuli. We conclude that there is currently no evidence supporting separate ON and OFF channels in stereopsis.

Binocular Summation for Reflexive Eye Movements: A Potential Diagnostic Tool for Stereodeficiencies.

Purpose: Stereoscopic vision, by detecting interocular correlations, enhances depth perception. Stereodeficiencies often emerge during the first months of life, and left untreated can lead to severe loss of visual acuity in one eye and/or strabismus. Early treatment results in much better outcomes, yet diagnostic tests for infants are cumbersome and not widely available. We asked whether reflexive eye movements, which in principle can be recorded even in infants, can be used to identify stereodeficiencies. Methods: Reflexive ocular following eye movements induced by fast drifting noise stimuli were recorded in 10 adult human participants (5 with normal stereoacuity, 5 stereodeficient). To manipulate interocular correlation, the stimuli shown to the two eyes were either identical, different, or had opposite contrast. Monocular presentations were also interleaved. The participants were asked to passively fixate the screen. Results: In the participants with normal stereoacuity, the responses to binocular identical stimuli were significantly larger than those induced by binocular opposite stimuli. In the stereodeficient participants the responses were indistinguishable. Despite the small size of ocular following responses, 40 trials, corresponding to less than 2 minutes of testing, were sufficient to reliably differentiate normal from stereodeficient participants. Conclusions: Ocular-following eye movements, because of their reliance on cortical neurons sensitive to interocular correlations, are affected by stereodeficiencies. Because these eye movements can be recorded noninvasively and with minimal participant cooperation, they can potentially be measured even in infants and might thus provide an useful screening tool for this currently underserved population.

Human infants can generate vergence responses to retinal disparity by 5 to 10 weeks of age.

Vergence is defined as a binocular eye movement during which the two eyes move in opposite directions to align to a target in depth. In adults, fine vergence control is driven primarily by interocular retinal image disparity. Although infants have not typically been shown to respond to disparity until 3 to 5 months postpartum, they have been shown to align their eyes from hours after birth. It remains unclear what drives these responses in young infants. In this experiment, 5- to 10-week-old human infants were presented with a dynamic random noise stimulus oscillating in disparity at 0.1 Hz over an amplitude of 2° for 30 s. Fourier transforms of the horizontal eye movements revealed significant disparity-driven responses at the frequency of the stimulus in over half of the tested infants. Because the stimulus updated dynamically, this experiment precluded the possibility of independent monocular fixations to a sustained target. These data demonstrate cortical binocular function in humans by five weeks, the youngest age tested here, which is as much as two months younger than previously believed.

Binocular summation for reflexive eye movements.

Psychophysical studies and our own subjective experience suggest that, in natural viewing conditions (i.e., at medium to high contrasts), monocularly and binocularly viewed scenes appear very similar, with the exception of the improved depth perception provided by stereopsis. This phenomenon is usually described as a lack of binocular summation. We show here that there is an exception to this rule: Ocular following eye movements induced by the sudden motion of a large stimulus, which we recorded from three human subjects, are much larger when both eyes see the moving stimulus, than when only one eye does. We further discovered that this binocular advantage is a function of the interocular correlation between the two monocular images: It is maximal when they are identical, and reduced when the two eyes are presented with different images. This is possible only if the neurons that underlie ocular following are sensitive to binocular disparity.

Feedback determines the structure of correlated variability in primary visual cortex.

The variable responses of sensory neurons tend to be weakly correlated (spike-count correlation, rsc). This is widely thought to reflect noise in shared afferents, in which case rsc can limit the reliability of sensory coding. However, it could also be due to feedback from higher-order brain regions. Currently, the relative contributions of these sources are unknown. We addressed this by recording from populations of V1 neurons in macaques performing different discrimination tasks involving the same visual input. We found that the structure of rsc (the way rsc varied with neuronal stimulus preference) changed systematically with task instruction. Therefore, even at the earliest stage in the cortical visual hierarchy, rsc structure during task performance primarily reflects feedback dynamics. Consequently, previous proposals for how rsc constrains sensory processing need not apply. Furthermore, we show that correlations between the activity of single neurons and choice depend on feedback engaged by the task.

Suppression and Contrast Normalization in Motion Processing.

Sensory neurons are activated by a range of stimuli to which they are said to be tuned. Usually, they are also suppressed by another set of stimuli that have little effect when presented in isolation. The interactions between preferred and suppressive stimuli are often quite complex and vary across neurons, even within a single area, making it difficult to infer their collective effect on behavioral responses mediated by activity across populations of neurons. Here, we investigated this issue by measuring, in human subjects (three males), the suppressive effect of static masks on the ocular following responses induced by moving stimuli. We found a wide range of effects, which depend in a nonlinear and nonseparable manner on the spatial frequency, contrast, and spatial location of both stimulus and mask. Under some conditions, the presence of the mask can be seen as scaling the contrast of the driving stimulus. Under other conditions, the effect is more complex, involving also a direct scaling of the behavioral response. All of this complexity at the behavioral level can be captured by a simple model in which stimulus and mask interact nonlinearly at two stages, one monocular and one binocular. The nature of the interactions is compatible with those observed at the level of single neurons in primates, usually broadly described as divisive normalization, without having to invoke any scaling mechanism.SIGNIFICANCE STATEMENT The response of sensory neurons to their preferred stimulus is often modulated by stimuli that are not effective when presented alone. Individual neurons can exhibit multiple modulatory effects, with considerable variability across neurons even in a single area. Such diversity has made it difficult to infer the impact of these modulatory mechanisms on behavioral responses. Here, we report the effects of a stationary mask on the reflexive eye movements induced by a moving stimulus. A model with two stages, each incorporating a divisive modulatory mechanism, reproduces our experimental results and suggests that qualitative variability of masking effects in cortical neurons might arise from differences in the extent to which such effects are inherited from earlier stages.

Visual Responses in FEF, Unlike V1, Primarily Reflect When the Visual Context Renders a Receptive Field Salient.

When light falls within a neuronal visual receptive field (RF) the resulting activity is referred to as the visual response. Recent work suggests this activity is in response to both the visual stimulation and the abrupt appearance, or salience, of the presentation. Here we present a novel method for distinguishing the two, based on the timing of random and nonrandom presentations. We examined these contributions in frontal eye field (FEF; N = 51) and as a comparison, an early stage in the primary visual cortex (V1; N = 15) of male monkeys (Macaca mulatta). An array of identical stimuli was presented within and outside the neuronal RF while we manipulated salience by varying the time between stimulus presentations. We hypothesized that the rapid presentation would reduce salience (the sudden appearance within the visual field) of a stimulus at any one location, and thus decrease responses driven by salience in the RF. We found that when the interstimulus interval decreased from 500 to 16 ms there was an approximate 79% reduction in the FEF response compared with an estimated 17% decrease in V1. This reduction in FEF response for rapid presentation was evident even when the random sequence preceding a stimulus did not stimulate the RF for 500 ms. The time course of these response changes in FEF suggest that salience is represented much earlier (<100 ms following stimulus onset) than previously estimated. Our results suggest that the contribution of salience dominates at higher levels of the visual system.SIGNIFICANCE STATEMENT The neuronal responses in early visual processing [e.g., primary visual cortex (V1)] reflect primarily the retinal stimulus. Processing in higher visual areas is modulated by a combination of the visual stimulation and contextual factors, such as salience, but identifying these components separately has been difficult. Here we quantified these contributions at a late stage of visual processing [frontal eye field (FEF)] and as a comparison, an early stage in V1. Our results suggest that as visual information continues through higher levels of processing the neural responses are no longer driven primarily by the visual stimulus in the receptive field, but by the broader context that stimulus defines-very different from current views about visual signals in FEF.

Visual Perception: Neural Networks for Stereopsis.

How does our brain use differences between the images in our two eyes, binocular disparities, to generate depth perception? New work shows that a type of neural network trained on natural binocular images can learn parameters that match key properties of visual cortex. Most information is conveyed by cells which sense differences between the two eyes' images.

Invisible noise obscures visible signal in insect motion detection.

The motion energy model is the standard account of motion detection in animals from beetles to humans. Despite this common basis, we show here that a difference in the early stages of visual processing between mammals and insects leads this model to make radically different behavioural predictions. In insects, early filtering is spatially lowpass, which makes the surprising prediction that motion detection can be impaired by "invisible" noise, i.e. noise at a spatial frequency that elicits no response when presented on its own as a signal. We confirm this prediction using the optomotor response of praying mantis Sphodromantis lineola. This does not occur in mammals, where spatially bandpass early filtering means that linear systems techniques, such as deriving channel sensitivity from masking functions, remain approximately valid. Counter-intuitive effects such as masking by invisible noise may occur in neural circuits wherever a nonlinearity is followed by a difference operation.

Combining 1-D components to extract pattern information: It is about more than component similarity.

At least under some conditions, plaid stimuli are processed by combining information first extracted in orientation and scale-selective channels. The rules that govern this combination across channels are only partially understood. Although the available data suggests that only components having similar spatial frequency and contrast are combined, the extent to which this holds has not been firmly established. To address this question, we measured, in human subjects, the short-latency reflexive vergence eye movements induced by stereo plaids in which spatial frequency and contrast of the components are independently varied. We found that, although similarity in component spatial frequency and contrast matter, they interact in a nonseparable way. One way in which this relationship might arise is if the internal estimate of contrast is not a faithful representation of stimulus contrast but is instead spatial frequency-dependent (with higher spatial frequencies being boosted). We propose that such weighting might have been put in place by a mechanism that, in an effort of achieve contrast constancy and/or coding efficiency, regulates the gain of detectors in early visual cortex to equalize their long-term average response to natural images.

Decision-Related Activity in Macaque V2 for Fine Disparity Discrimination Is Not Compatible with Optimal Linear Readout.

Fine judgments of stereoscopic depth rely mainly on relative judgments of depth (relative binocular disparity) between objects, rather than judgments of the distance to where the eyes are fixating (absolute disparity). In macaques, visual area V2 is the earliest site in the visual processing hierarchy for which neurons selective for relative disparity have been observed (Thomas et al., 2002). Here, we found that, in macaques trained to perform a fine disparity discrimination task, disparity-selective neurons in V2 were highly selective for the task, and their activity correlated with the animals' perceptual decisions (unexplained by the stimulus). This may partially explain similar correlations reported in downstream areas. Although compatible with a perceptual role of these neurons for the task, the interpretation of such decision-related activity is complicated by the effects of interneuronal "noise" correlations between sensory neurons. Recent work has developed simple predictions to differentiate decoding schemes (Pitkow et al., 2015) without needing measures of noise correlations, and found that data from early sensory areas were compatible with optimal linear readout of populations with information-limiting correlations. In contrast, our data here deviated significantly from these predictions. We additionally tested this prediction for previously reported results of decision-related activity in V2 for a related task, coarse disparity discrimination (Nienborg and Cumming, 2006), thought to rely on absolute disparity. Although these data followed the predicted pattern, they violated the prediction quantitatively. This suggests that optimal linear decoding of sensory signals is not generally a good predictor of behavior in simple perceptual tasks. SIGNIFICANCE STATEMENT: Activity in sensory neurons that correlates with an animal's decision is widely believed to provide insights into how the brain uses information from sensory neurons. Recent theoretical work developed simple predictions to differentiate decoding schemes, and found support for optimal linear readout of early sensory populations with information-limiting correlations. Here, we observed decision-related activity for neurons in visual area V2 of macaques performing fine disparity discrimination, as yet the earliest site for this task. These findings, and previously reported results from V2 in a different task, deviated from the predictions for optimal linear readout of a population with information-limiting correlations. Our results suggest that optimal linear decoding of early sensory information is not a general decoding strategy used by the brain.