Discriminating Audiovisual Speed: Optimal Integration of Speed Defaults to Probability Summation When Component Reliabilities Diverge

Statistically Optimal Cue Integration During Human Spatial Navigation

In 2007, Cheng and colleagues published their influential review wherein they analyzed the literature on spatial cue interaction during navigation through a Bayesian lens, and concluded that models of optimal cue integration often applied in psychophysical studies could explain cue interaction during navigation. Since then, numerous empirical investigations have been conducted to assess the degree to which human navigators are optimal when integrating multiple spatial cues during a variety of navigation-related tasks. In the current review, we discuss the literature on human cue integration during navigation that has been published since Cheng et al.'s original review. Evidence from most studies demonstrate optimal navigation behavior when humans are presented with multiple spatial cues. However, applications of optimal cue integration models vary in their underlying assumptions (e.g., uninformative priors and decision rules). Furthermore, cue integration behavior depends in part on the nature of the cues being integrated and the navigational task (e.g., homing versus non-home goal localization). We discuss the implications of these models and suggest directions for future research.

Perceived speed at low luminance: Lights out for the Bayesian observer?

To account for perceptual bias, Bayesian models use the precision of early sensory measurements to weight the influence of prior expectations. As precision decreases, prior expectations start to dominate. Important examples come from motion perception, where the slow-motion prior has been used to explain a variety of motion illusions in vision, hearing, and touch, many of which correlate appropriately with threshold measures of underlying precision. However, the Bayesian account seems defeated by the finding that moving objects appear faster in the dark, because most motion thresholds are worse at low luminance. Here we show this is not the case for speed discrimination. Our results show that performance improves at low light levels by virtue of a perceived contrast cue that is more salient in the dark. With this cue removed, discrimination becomes independent of luminance. However, we found perceived speed still increased in the dark for the same observers, and by the same amount. A possible interpretation is that motion processing is therefore not Bayesian, because our findings challenge a key assumption these models make, namely that the accuracy of early sensory measurements is independent of basic stimulus properties like luminance. However, a final experiment restored Bayesian behaviour by adding external noise, making discrimination worse and slowing perceived speed down. Our findings therefore suggest that motion is processed in a Bayesian fashion but based on noisy sensory measurements that also vary in accuracy.

A review of interactions between peripheral and foveal vision

Visual processing varies dramatically across the visual field. These differences start in the retina and continue all the way to the visual cortex. Despite these differences in processing, the perceptual experience of humans is remarkably stable and continuous across the visual field. Research in the last decade has shown that processing in peripheral and foveal vision is not independent, but is more directly connected than previously thought. We address three core questions on how peripheral and foveal vision interact, and review recent findings on potentially related phenomena that could provide answers to these questions. First, how is the processing of peripheral and foveal signals related during fixation? Peripheral signals seem to be processed in foveal retinotopic areas to facilitate peripheral object recognition, and foveal information seems to be extrapolated toward the periphery to generate a homogeneous representation of the environment. Second, how are peripheral and foveal signals re-calibrated? Transsaccadic changes in object features lead to a reduction in the discrepancy between peripheral and foveal appearance. Third, how is peripheral and foveal information stitched together across saccades? Peripheral and foveal signals are integrated across saccadic eye movements to average percepts and to reduce uncertainty. Together, these findings illustrate that peripheral and foveal processing are closely connected, mastering the compromise between a large peripheral visual field and high resolution at the fovea.

Ageing and multisensory integration: A review of the evidence, and a computational perspective

The processing of multisensory signals is crucial for effective interaction with the environment, but our ability to perform this vital function changes as we age. In the first part of this review, we summarise existing research into the effects of healthy ageing on multisensory integration. We note that age differences vary substantially with the paradigms and stimuli used: older adults often receive at least as much benefit (to both accuracy and response times) as younger controls from congruent multisensory stimuli, but are also consistently more negatively impacted by the presence of intersensory conflict. In the second part, we outline a normative Bayesian framework that provides a principled and computationally informed perspective on the key ingredients involved in multisensory perception, and how these are affected by ageing. Applying this framework to the existing literature, we conclude that changes to sensory reliability, prior expectations (together with attentional control), and decisional strategies all contribute to the age differences observed. However, we find no compelling evidence of any age-related changes to the basic inference mechanisms involved in multisensory perception.

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.

Integration of audiovisual spatial signals is not consistent with maximum likelihood estimation

Multisensory perception is regarded as one of the most prominent examples where human behaviour conforms to the computational principles of maximum likelihood estimation (MLE). In particular, observers are thought to integrate auditory and visual spatial cues weighted in proportion to their relative sensory reliabilities into the most reliable and unbiased percept consistent with MLE. Yet, evidence to date has been inconsistent. The current pre-registered, large-scale (N = 36) replication study investigated the extent to which human behaviour for audiovisual localization is in line with maximum likelihood estimation. The acquired psychophysics data show that while observers were able to reduce their multisensory variance relative to the unisensory variances in accordance with MLE, they weighed the visual signals significantly stronger than predicted by MLE. Simulations show that this dissociation can be explained by a greater sensitivity of standard estimation procedures to detect deviations from MLE predictions for sensory weights than for audiovisual variances. Our results therefore suggest that observers did not integrate audiovisual spatial signals weighted exactly in proportion to their relative reliabilities for localization. These small deviations from the predictions of maximum likelihood estimation may be explained by observers' uncertainty about the world's causal structure as accounted for by Bayesian causal inference.

Optimal trans-saccadic integration relies on visual working memory

Saccadic eye movements alter the visual processing of objects of interest by bringing them from the periphery, where there is only low-resolution vision, to the high-resolution fovea. Evidence suggests that people are able to achieve trans-saccadic integration in a near-optimal manner; however the mechanisms underlying integration are still unclear. Visual working memory (VWM) is sustained across a saccade, and it has been suggested that this memory resource is used to store and compare the pre- and post- saccadic percepts. This study directly tested the hypothesis that VWM is necessary for optimal trans-saccadic integration, by introducing memory load during a saccade, and testing subsequent integration performance on feature similar and dissimilar stimuli. Results show that integration performance was impaired when there was an additional memory task. Additionally, performance on the memory task was affected by feature-specific integration stimuli. Our results suggest that VWM supports the integration of pre- and post- saccadic stimuli because integration performance is impaired under VWM load.

Computational characterization of visually induced auditory spatial adaptation

Recent research investigating the principles governing human perception has provided increasing evidence for probabilistic inference in human perception. For example, human auditory and visual localization judgments closely resemble that of a Bayesian causal inference observer, where the underlying causal structure of the stimuli are inferred based on both the available sensory evidence and prior knowledge. However, most previous studies have focused on characterization of perceptual inference within a static environment, and therefore, little is known about how this inference process changes when observers are exposed to a new environment. In this study we aimed to computationally characterize the change in auditory spatial perception induced by repeated auditory–visual spatial conflict, known as the ventriloquist aftereffect. In theory, this change could reflect a shift in the auditory sensory representations (i.e., shift in auditory likelihood distribution), a decrease in the precision of the auditory estimates (i.e., increase in spread of likelihood distribution), a shift in the auditory bias (i.e., shift in prior distribution), or an increase/decrease in strength of the auditory bias (i.e., the spread of prior distribution), or a combination of these. By quantitatively estimating the parameters of the perceptual process for each individual observer using a Bayesian causal inference model, we found that the shift in the perceived locations after exposure was associated with a shift in the mean of the auditory likelihood functions in the direction of the experienced visual offset. The results suggest that repeated exposure to a fixed auditory–visual discrepancy is attributed by the nervous system to sensory representation error and as a result, the sensory map of space is recalibrated to correct the error.

Integration of vestibular and proprioceptive signals for spatial updating

Spatial updating during self-motion typically involves the appropriate integration of both visual and non-visual cues, including vestibular and proprioceptive information. Here, we investigated how human observers combine these two non-visual cues during full-stride curvilinear walking. To obtain a continuous, real-time estimate of perceived position, observers were asked to continuously point toward a previously viewed target in the absence of vision. They did so while moving on a large circular treadmill under various movement conditions. Two conditions were designed to evaluate spatial updating when information was largely limited to either proprioceptive information (walking in place) or vestibular information (passive movement). A third condition evaluated updating when both sources of information were available (walking through space) and were either congruent or in conflict. During both the passive movement condition and while walking through space, the pattern of pointing behavior demonstrated evidence of accurate egocentric updating. In contrast, when walking in place, perceived self-motion was underestimated and participants always adjusted the pointer at a constant rate, irrespective of changes in the rate at which the participant moved relative to the target. The results are discussed in relation to the maximum likelihood estimation model of sensory integration. They show that when the two cues were congruent, estimates were combined, such that the variance of the adjustments was generally reduced. Results also suggest that when conflicts were introduced between the vestibular and proprioceptive cues, spatial updating was based on a weighted average of the two inputs.