How multisensory neurons solve causal inference.

Sitting in a static railway carriage can produce illusory self-motion if the train on an adjoining track moves off. While our visual system registers motion, vestibular signals indicate that we are stationary. The brain is faced with a difficult challenge: is there a single cause of sensations (I am moving) or two causes (I am static, another train is moving)? If a single cause, integrating signals produces a more precise estimate of self-motion, but if not, one cue should be ignored. In many cases, this process of causal inference works without error, but how does the brain achieve it? Electrophysiological recordings show that the macaque medial superior temporal area contains many neurons that encode combinations of vestibular and visual motion cues. Some respond best to vestibular and visual motion in the same direction ("congruent" neurons), while others prefer opposing directions ("opposite" neurons). Congruent neurons could underlie cue integration, but the function of opposite neurons remains a puzzle. Here, we seek to explain this computational arrangement by training a neural network model to solve causal inference for motion estimation. Like biological systems, the model develops congruent and opposite units and recapitulates known behavioral and neurophysiological observations. We show that all units (both congruent and opposite) contribute to motion estimation. Importantly, however, it is the balance between their activity that distinguishes whether visual and vestibular cues should be integrated or separated. This explains the computational purpose of puzzling neural representations and shows how a relatively simple feedforward network can solve causal inference.

Ultra-High-Field Neuroimaging Reveals Fine-Scale Processing for 3D Perception.

Binocular disparity provides critical information about three-dimensional (3D) structures to support perception and action. In the past decade significant progress has been made in uncovering human brain areas engaged in the processing of binocular disparity signals. Yet, the fine-scale brain processing underlying 3D perception remains unknown. Here, we use ultra-high-field (7T) functional imaging at submillimeter resolution to examine fine-scale BOLD fMRI signals involved in 3D perception. In particular, we sought to interrogate the local circuitry involved in disparity processing by sampling fMRI responses at different positions relative to the cortical surface (i.e., across cortical depths corresponding to layers). We tested for representations related to 3D perception by presenting participants (male and female, N = 8) with stimuli that enable stable stereoscopic perception [i.e., correlated random dot stereograms (RDS)] versus those that do not (i.e., anticorrelated RDS). Using multivoxel pattern analysis (MVPA), we demonstrate cortical depth-specific representations in areas V3A and V7 as indicated by stronger pattern responses for correlated than for anticorrelated stimuli in upper rather than deeper layers. Examining informational connectivity, we find higher feedforward layer-to-layer connectivity for correlated than anticorrelated stimuli between V3A and V7. Further, we observe disparity-specific feedback from V3A to V1 and from V7 to V3A. Our findings provide evidence for the role of V3A as a key nexus for disparity processing, which is implicated in feedforward and feedback signals related to the perceptual estimation of 3D structures.SIGNIFICANCE STATEMENT Binocular vision plays a significant role in supporting our interactions with the surrounding environment. The fine-scale neural mechanisms that underlie the brain's skill in extracting 3D structures from binocular signals are poorly understood. Here, we capitalize on recent advances in ultra-high-field functional imaging to interrogate human brain circuits involved in 3D perception at submillimeter resolution. We provide evidence for the role of area V3A as a key nexus for disparity processing, which is implicated in feedforward and feedback signals related to the perceptual estimation of 3D structures from binocular signals. These fine-scale measurements help bridge the gap between animal neurophysiology and human fMRI studies investigating cross-scale circuits, from micro circuits to global brain networks for 3D perception.

Exploring and explaining properties of motion processing in biological brains using a neural network.

Visual motion perception underpins behaviors ranging from navigation to depth perception and grasping. Our limited access to biological systems constrains our understanding of how motion is processed within the brain. Here we explore properties of motion perception in biological systems by training a neural network to estimate the velocity of image sequences. The network recapitulates key characteristics of motion processing in biological brains, and we use our access to its structure to explore and understand motion (mis)perception. We find that the network captures the biological response to reverse-phi motion in terms of direction. We further find that it overestimates and underestimates the speed of slow and fast reverse-phi motion, respectively, because of the correlation between reverse-phi motion and the spatiotemporal receptive fields tuned to motion in opposite directions. Second, we find that the distribution of spatiotemporal tuning properties in the V1 and middle temporal (MT) layers of the network are similar to those observed in biological systems. We then show that, in comparison to MT units tuned to fast speeds, those tuned to slow speeds primarily receive input from V1 units tuned to high spatial frequency and low temporal frequency. Next, we find that there is a positive correlation between the pattern-motion and speed selectivity of MT units. Finally, we show that the network captures human underestimation of low coherence motion stimuli, and that this is due to pooling of noise and signal motion. These findings provide biologically plausible explanations for well-known phenomena and produce concrete predictions for future psychophysical and neurophysiological experiments.

But Still It Moves: Static Image Statistics Underlie How We See Motion.

Seeing movement promotes survival. It results from an uncertain interplay between evolution and experience, making it hard to isolate the drivers of computational architectures found in brains. Here we seek insight into motion perception using a neural network (MotionNet) trained on moving images to classify velocity. The network recapitulates key properties of motion direction and speed processing in biological brains, and we use it to derive, and test, understanding of motion (mis)perception at the computational, neural, and perceptual levels. We show that diverse motion characteristics are largely explained by the statistical structure of natural images, rather than motion per se. First, we show how neural and perceptual biases for particular motion directions can result from the orientation structure of natural images. Second, we demonstrate an interrelation between speed and direction preferences in (macaque) MT neurons that can be explained by image autocorrelation. Third, we show that natural image statistics mean that speed and image contrast are related quantities. Finally, using behavioral tests (humans, both sexes), we show that it is knowledge of the speed-contrast association that accounts for motion illusions, rather than the distribution of movements in the environment (the "slow world" prior) as premised by Bayesian accounts. Together, this provides an exposition of motion speed and direction estimation, and produces concrete predictions for future neurophysiological experiments. More broadly, we demonstrate the conceptual value of marrying artificial systems with biological characterization, moving beyond "black box" reproduction of an architecture to advance understanding of complex systems, such as the brain.SIGNIFICANCE STATEMENT Using an artificial systems approach, we show that physiological properties of motion can result from natural image structure. In particular, we show that the anisotropic distribution of orientations in natural statistics is sufficient to explain the cardinal bias for motion direction. We show that inherent autocorrelation in natural images means that speed and direction are related quantities, which could shape the relationship between speed and direction tuning of MT neurons. Finally, we show that movement speed and image contrast are related in moving natural images, and that motion misperception can be explained by this speed-contrast association not a "slow world" prior.

Adaptation to Binocular Anticorrelation Results in Increased Neural Excitability.

Throughout the brain, information from individual sources converges onto higher order neurons. For example, information from the two eyes first converges in binocular neurons in area V1. Some neurons are tuned to similarities between sources of information, which makes intuitive sense in a system striving to match multiple sensory signals to a single external cause-that is, establish causal inference. However, there are also neurons that are tuned to dissimilar information. In particular, some binocular neurons respond maximally to a dark feature in one eye and a light feature in the other. Despite compelling neurophysiological and behavioral evidence supporting the existence of these neurons [Katyal, S., Vergeer, M., He, S., He, B., & Engel, S. A. Conflict-sensitive neurons gate interocular suppression in human visual cortex. Scientific Reports, 8, 1239, 2018; Kingdom, F. A. A., Jennings, B. J., & Georgeson, M. A. Adaptation to interocular difference. Journal of Vision, 18, 9, 2018; Janssen, P., Vogels, R., Liu, Y., & Orban, G. A. At least at the level of inferior temporal cortex, the stereo correspondence problem is solved. Neuron, 37, 693-701, 2003; Tsao, D. Y., Conway, B. R., & Livingstone, M. S. Receptive fields of disparity-tuned simple cells in macaque V1. Neuron, 38, 103-114, 2003; Cumming, B. G., & Parker, A. J. Responses of primary visual cortical neurons to binocular disparity without depth perception. Nature, 389, 280-283, 1997], their function has remained opaque. To determine how neural mechanisms tuned to dissimilarities support perception, here we use electroencephalography to measure human observers' steady-state visually evoked potentials in response to change in depth after prolonged viewing of anticorrelated and correlated random-dot stereograms (RDS). We find that adaptation to anticorrelated RDS results in larger steady-state visually evoked potentials, whereas adaptation to correlated RDS has no effect. These results are consistent with recent theoretical work suggesting "what not" neurons play a suppressive role in supporting stereopsis [Goncalves, N. R., & Welchman, A. E. "What not" detectors help the brain see in depth. Current Biology, 27, 1403-1412, 2017]; that is, selective adaptation of neurons tuned to binocular mismatches reduces suppression resulting in increased neural excitability.

Learning predictive structure without a teacher: decision strategies and brain routes.

Extracting the structure of complex environments is at the core of our ability to interpret the present and predict the future. This skill is important for a range of behaviours from navigating a new city to learning music and language. Classical approaches that investigate our ability to extract the principles of organisation that govern complex environments focus on reward-based learning. Yet, the human brain is shown to be expert at learning generative structure based on mere exposure and without explicit reward. Individuals are shown to adapt to-unbeknownst to them-changes in the environment's temporal statistics and predict future events. Further, we present evidence for a common brain architecture for unsupervised structure learning and reward-based learning, suggesting that the brain is built on the premise that 'learning is its own reward' to support adaptive behaviour.

Mixed-polarity random-dot stereograms alter GABA and Glx concentration in the early visual cortex.

The offset between images projected onto the left and right retina (binocular disparity) provides a powerful cue to the three-dimensional structure of the environment. It was previously shown that depth judgements are better when images comprise both light and dark features, rather than only light or only dark elements. Since Harris and Parker (Nature 374: 808-811, 1995) discovered the "mixed-polarity benefit," there has been limited evidence supporting their hypothesis that the benefit is due to separate bright and dark channels. Goncalves and Welchman (Curr Biol 27: 1403-1412, 2017) observed that single- and mixed-polarity stereograms evoke different levels of positive and negative activity in a deep neural network trained on natural images to make depth judgements, which also showed the mixed-polarity benefit. Motivated by this discovery, we seek to test the potential for changes in the balance of excitation and inhibition that are produced by viewing these stimuli. In particular, we use magnetic resonance spectroscopy to measure Glx and GABA concentrations in the early visual cortex of adult humans during viewing of single- and mixed-polarity random-dot stereograms (RDS). We find that participants' Glx concentration is significantly higher, whereas GABA concentration is significantly lower, when mixed-polarity RDS are viewed than when single-polarity RDS are viewed. These results indicate that excitation and inhibition facilitate processing of single- and mixed-polarity stereograms in the early visual cortex to different extents, consistent with recent theoretical work (Goncalves NR, Welchman AE. Curr Biol 27: 1403-1412, 2017).NEW & NOTEWORTHY Depth judgements are better when images comprise both light and dark features, rather than only light or only dark elements. Using magnetic resonance spectroscopy, we show that adult human participants' Glx concentration is significantly higher whereas GABA concentration is significantly lower in the early visual cortex when participants view mixed-polarity random-dot stereograms (RDS) compared with single-polarity RDS. These results indicate that excitation and inhibition facilitate processing of single- and mixed-polarity stereograms in the early visual cortex to different extents.

Contextual effects on binocular matching are evident in primary visual cortex.

Global context can dramatically influence local visual perception. This phenomenon is well-documented for monocular features, e.g., the Kanizsa triangle. It has been demonstrated for binocular matching: the disambiguation of the Wallpaper Illusion via the luminance of the background. For monocular features, there is evidence that global context can influence neuronal responses as early as V1. However, for binocular matching, the activity in this area of the visual cortex is thought to represent local processing, suggesting that the influence of global context may occur at later stages of cortical processing. Here we sought to test if binocular matching is influenced by contextual effects in V1, using fMRI to measure brain activity while participants viewed perceptually ambiguous "wallpaper" stereograms whose depth was disambiguated by the luminance of the surrounding region. We localized voxels in V1 corresponding to the ambiguous region of the pattern, i.e., where the signal received from the eyes was not predictive of depth, and despite the ambiguity of the input signal, using multi-voxel pattern analysis we were able to reliably decode perceived (near/far) depth from the activity of these voxels. These findings indicate that stereoscopic related neural activity is influenced by global context as early as V1.

Areal differences in depth cue integration between monkey and human.

Electrophysiological evidence suggested primarily the involvement of the middle temporal (MT) area in depth cue integration in macaques, as opposed to human imaging data pinpointing area V3B/kinetic occipital area (V3B/KO). To clarify this conundrum, we decoded monkey functional MRI (fMRI) responses evoked by stimuli signaling near or far depths defined by binocular disparity, relative motion, and their combination, and we compared results with those from an identical experiment previously performed in humans. Responses in macaque area MT are more discriminable when two cues concurrently signal depth, and information provided by one cue is diagnostic of depth indicated by the other. This suggests that monkey area MT computes fusion of disparity and motion depth signals, exactly as shown for human area V3B/KO. Hence, these data reconcile previously reported discrepancies between depth processing in human and monkey by showing the involvement of the dorsal stream in depth cue integration using the same technique, despite the engagement of different regions.

Multimodal imaging of brain connectivity reveals predictors of individual decision strategy in statistical learning.

Successful human behaviour depends on the brain's ability to extract meaningful structure from information streams and make predictions about future events. Individuals can differ markedly in the decision strategies they use to learn the environment's statistics, yet we have little idea why. Here, we investigate whether the brain networks involved in learning temporal sequences without explicit reward differ depending on the decision strategy that individuals adopt. We demonstrate that individuals alter their decision strategy in response to changes in temporal statistics and engage dissociable circuits: extracting the exact sequence statistics relates to plasticity in motor corticostriatal circuits, while selecting the most probable outcomes relates to plasticity in visual, motivational and executive corticostriatal circuits. Combining graph metrics of functional and structural connectivity, we provide evidence that learning-dependent changes in these circuits predict individual decision strategy. Our findings propose brain plasticity mechanisms that mediate individual ability for interpreting the structure of variable environments.

The mixed-polarity benefit of stereopsis arises in early visual cortex.

Depth perception is better when observers view stimuli containing a mixture of bright and dark visual features. It is currently unclear where in the visual system sensory processing benefits from the availability of different contrast polarity. To address this question, we applied transcranial magnetic stimulation to the visual cortex to modulate normal neural activity during processing of single- or mixed-polarity random-dot stereograms. In line with previous work, participants gave significantly better depth judgments for mixed-polarity stimuli. Stimulation of early visual cortex (V1/V2) significantly increased this benefit for mixed-polarity stimuli, and it did not affect performance for single-polarity stimuli. Stimulation of disparity responsive areas V3a and LO had no effect on perception. Our findings show that disparity processing in early visual cortex gives rise to the mixed-polarity benefit. This is consistent with computational models of stereopsis at the level of V1 that produce a mixed polarity benefit.

Shape Perception: Boundary Conditions on a Grey Area.

Modulations in light intensity across a visual image could be caused by a flat object with varying pigmentation, such as wallpaper, or differential light reflection from a three-dimensional shape made of uniform material, such as curtains. A new study identifies key image cues that help the brain work out which interpretation to select.

White-Matter Pathways for Statistical Learning of Temporal Structures.

Extracting the statistics of event streams in natural environments is critical for interpreting current events and predicting future ones. The brain is known to rapidly find structure and meaning in unfamiliar streams of sensory experience, often by mere exposure to the environment (i.e., without explicit feedback). Yet, we know little about the brain pathways that support this type of statistical learning. Here, we test whether changes in white-matter (WM) connectivity due to training relate to our ability to extract temporal regularities. By combining behavioral training and diffusion tensor imaging (DTI), we demonstrate that humans adapt to the environment's statistics as they change over time from simple repetition to probabilistic combinations. In particular, we show that learning relates to the decision strategy that individuals adopt when extracting temporal statistics. We next test for learning-dependent changes in WM connectivity and ask whether they relate to individual variability in decision strategy. Our DTI results provide evidence for dissociable WM pathways that relate to individual strategy: extracting the exact sequence statistics (i.e., matching) relates to connectivity changes between caudate and hippocampus, while selecting the most probable outcomes in a given context (i.e., maximizing) relates to connectivity changes between prefrontal, cingulate and basal ganglia (caudate, putamen) regions. Thus, our findings provide evidence for distinct cortico-striatal circuits that show learning-dependent changes of WM connectivity and support individual ability to learn behaviorally-relevant statistics.

Proscription supports robust perceptual integration by suppression in human visual cortex.

Perception relies on integrating information within and between the senses, but how does the brain decide which pieces of information should be integrated and which kept separate? Here we demonstrate how proscription can be used to solve this problem: certain neurons respond best to unrealistic combinations of features to provide 'what not' information that drives suppression of unlikely perceptual interpretations. First, we present a model that captures both improved perception when signals are consistent (and thus should be integrated) and robust estimation when signals are conflicting. Second, we test for signatures of proscription in the human brain. We show that concentrations of inhibitory neurotransmitter GABA in a brain region intricately involved in integrating cues (V3B/KO) correlate with robust integration. Finally, we show that perturbing excitation/inhibition impairs integration. These results highlight the role of proscription in robust perception and demonstrate the functional purpose of 'what not' sensors in supporting sensory estimation.

Learning predictive statistics from temporal sequences: Dynamics and strategies.

Human behavior is guided by our expectations about the future. Often, we make predictions by monitoring how event sequences unfold, even though such sequences may appear incomprehensible. Event structures in the natural environment typically vary in complexity, from simple repetition to complex probabilistic combinations. How do we learn these structures? Here we investigate the dynamics of structure learning by tracking human responses to temporal sequences that change in structure unbeknownst to the participants. Participants were asked to predict the upcoming item following a probabilistic sequence of symbols. Using a Markov process, we created a family of sequences, from simple frequency statistics (e.g., some symbols are more probable than others) to context-based statistics (e.g., symbol probability is contingent on preceding symbols). We demonstrate the dynamics with which individuals adapt to changes in the environment's statistics-that is, they extract the behaviorally relevant structures to make predictions about upcoming events. Further, we show that this structure learning relates to individual decision strategy; faster learning of complex structures relates to selection of the most probable outcome in a given context (maximizing) rather than matching of the exact sequence statistics. Our findings provide evidence for alternate routes to learning of behaviorally relevant statistics that facilitate our ability to predict future events in variable environments.

Learning Predictive Statistics: Strategies and Brain Mechanisms.

When immersed in a new environment, we are challenged to decipher initially incomprehensible streams of sensory information. However, quite rapidly, the brain finds structure and meaning in these incoming signals, helping us to predict and prepare ourselves for future actions. This skill relies on extracting the statistics of event streams in the environment that contain regularities of variable complexity from simple repetitive patterns to complex probabilistic combinations. Here, we test the brain mechanisms that mediate our ability to adapt to the environment's statistics and predict upcoming events. By combining behavioral training and multisession fMRI in human participants (male and female), we track the corticostriatal mechanisms that mediate learning of temporal sequences as they change in structure complexity. We show that learning of predictive structures relates to individual decision strategy; that is, selecting the most probable outcome in a given context (maximizing) versus matching the exact sequence statistics. These strategies engage distinct human brain regions: maximizing engages dorsolateral prefrontal, cingulate, sensory-motor regions, and basal ganglia (dorsal caudate, putamen), whereas matching engages occipitotemporal regions (including the hippocampus) and basal ganglia (ventral caudate). Our findings provide evidence for distinct corticostriatal mechanisms that facilitate our ability to extract behaviorally relevant statistics to make predictions.SIGNIFICANCE STATEMENT Making predictions about future events relies on interpreting streams of information that may initially appear incomprehensible. Past work has studied how humans identify repetitive patterns and associative pairings. However, the natural environment contains regularities that vary in complexity from simple repetition to complex probabilistic combinations. Here, we combine behavior and multisession fMRI to track the brain mechanisms that mediate our ability to adapt to changes in the environment's statistics. We provide evidence for an alternate route for learning complex temporal statistics: extracting the most probable outcome in a given context is implemented by interactions between executive and motor corticostriatal mechanisms compared with visual corticostriatal circuits (including hippocampal cortex) that support learning of the exact temporal statistics.

"What Not" Detectors Help the Brain See in Depth.

Binocular stereopsis is one of the primary cues for three-dimensional (3D) vision in species ranging from insects to primates. Understanding how the brain extracts depth from two different retinal images represents a tractable challenge in sensory neuroscience that has so far evaded full explanation. Central to current thinking is the idea that the brain needs to identify matching features in the two retinal images (i.e., solving the "stereoscopic correspondence problem") so that the depth of objects in the world can be triangulated. Although intuitive, this approach fails to account for key physiological and perceptual observations. We show that formulating the problem to identify "correct matches" is suboptimal and propose an alternative, based on optimal information encoding, that mixes disparity detection with "proscription": exploiting dissimilar features to provide evidence against unlikely interpretations. We demonstrate the role of these "what not" responses in a neural network optimized to extract depth in natural images. The network combines information for and against the likely depth structure of the viewed scene, naturally reproducing key characteristics of both neural responses and perceptual interpretations. We capture the encoding and readout computations of the network in simple analytical form and derive a binocular likelihood model that provides a unified account of long-standing puzzles in 3D vision at the physiological and perceptual levels. We suggest that marrying detection with proscription provides an effective coding strategy for sensory estimation that may be useful for diverse feature domains (e.g., motion) and multisensory integration.

Mapping the visual brain areas susceptible to phosphene induction through brain stimulation.

Transcranial magnetic stimulation (TMS) is a non-invasive brain stimulation technique whose effects on neural activity can be uncertain. Within the visual cortex, phosphenes are a useful marker of TMS: They indicate the induction of neural activation that propagates and creates a conscious percept. However, we currently do not know how susceptible different areas of the visual cortex are to TMS-induced phosphenes. In this study, we systematically map out locations in the visual cortex where stimulation triggered phosphenes. We relate this to the retinotopic organization and the location of object- and motion-selective areas, identified by functional magnetic resonance imaging (fMRI) measurements. Our results show that TMS can reliably induce phosphenes in early (V1, V2d, and V2v) and dorsal (V3d and V3a) visual areas close to the interhemispheric cleft. However, phosphenes are less likely in more lateral locations (hMT+/V5 and LOC). This suggests that early and dorsal visual areas are particularly amenable to TMS and that TMS can be used to probe the functional role of these areas.

Lesions to right posterior parietal cortex impair visual depth perception from disparity but not motion cues.

The posterior parietal cortex (PPC) is understood to be active when observers perceive three-dimensional (3D) structure. However, it is not clear how central this activity is in the construction of 3D spatial representations. Here, we examine whether PPC is essential for two aspects of visual depth perception by testing patients with lesions affecting this region. First, we measured subjects' ability to discriminate depth structure in various 3D surfaces and objects using binocular disparity. Patients with lesions to right PPC (N = 3) exhibited marked perceptual deficits on these tasks, whereas those with left hemisphere lesions (N = 2) were able to reliably discriminate depth as accurately as control subjects. Second, we presented an ambiguous 3D stimulus defined by structure from motion to determine whether PPC lesions influence the rate of bistable perceptual alternations. Patients' percept durations for the 3D stimulus were generally within a normal range, although the two patients with bilateral PPC lesions showed the fastest perceptual alternation rates in our sample. Intermittent stimulus presentation reduced the reversal rate similarly across subjects. Together, the results suggest that PPC plays a causal role in both inferring and maintaining the perception of 3D structure with stereopsis supported primarily by the right hemisphere, but do not lend support to the view that PPC is a critical contributor to bistable perceptual alternations.This article is part of the themed issue 'Vision in our three-dimensional world'.