Bayesian integration of visual and auditory signals for spatial localization

Bayesian Analysis of Perceived Eye Level

To accurately perceive the world, people must efficiently combine internal beliefs and external sensory cues. We introduce a Bayesian framework that explains the role of internal balance cues and visual stimuli on perceived eye level (PEL)—a self-reported measure of elevation angle. This framework provides a single, coherent model explaining a set of experimentally observed PEL over a range of experimental conditions. Further, it provides a parsimonious explanation for the additive effect of low fidelity cues as well as the averaging effect of high fidelity cues, as also found in other Bayesian cue combination psychophysical studies. Our model accurately estimates the PEL and explains the form of previous equations used in describing PEL behavior. Most importantly, the proposed Bayesian framework for PEL is more powerful than previous behavioral modeling; it permits behavioral estimation in a wider range of cue combination and perceptual studies than models previously reported.

Modeling the Perception of Audiovisual Distance: Bayesian Causal Inference and Other Models

Studies of audiovisual perception of distance are rare. Here, visual and auditory cue interactions in distance are tested against several multisensory models, including a modified causal inference model. In this causal inference model predictions of estimate distributions are included. In our study, the audiovisual perception of distance was overall better explained by Bayesian causal inference than by other traditional models, such as sensory dominance and mandatory integration, and no interaction. Causal inference resolved with probability matching yielded the best fit to the data. Finally, we propose that sensory weights can also be estimated from causal inference. The analysis of the sensory weights allows us to obtain windows within which there is an interaction between the audiovisual stimuli. We find that the visual stimulus always contributes by more than 80% to the perception of visual distance. The visual stimulus also contributes by more than 50% to the perception of auditory distance, but only within a mobile window of interaction, which ranges from 1 to 4 m.

Comparison of congruence judgment and auditory localization tasks for assessing the spatial limits of visual capture

Vision typically has better spatial accuracy and precision than audition and as a result often captures auditory spatial perception when visual and auditory cues are presented together. One determinant of visual capture is the amount of spatial disparity between auditory and visual cues: when disparity is small, visual capture is likely to occur, and when disparity is large, visual capture is unlikely. Previous experiments have used two methods to probe how visual capture varies with spatial disparity. First, congruence judgment assesses perceived unity between cues by having subjects report whether or not auditory and visual targets came from the same location. Second, auditory localization assesses the graded influence of vision on auditory spatial perception by having subjects point to the remembered location of an auditory target presented with a visual target. Previous research has shown that when both tasks are performed concurrently they produce similar measures of visual capture, but this may not hold when tasks are performed independently. Here, subjects alternated between tasks independently across three sessions. A Bayesian inference model of visual capture was used to estimate perceptual parameters for each session, which were compared across tasks. Results demonstrated that the range of audiovisual disparities over which visual capture was likely to occur was narrower in auditory localization than in congruence judgment, which the model indicates was caused by subjects adjusting their prior expectation that targets originated from the same location in a task-dependent manner.

Accumulation and decay of visual capture and the ventriloquism aftereffect caused by brief audio-visual disparities

Visual capture and the ventriloquism aftereffect resolve spatial disparities of incongruent auditory visual (AV) objects by shifting auditory spatial perception to align with vision. Here, we demonstrated the distinct temporal characteristics of visual capture and the ventriloquism aftereffect in response to brief AV disparities. In a set of experiments, subjects localized either the auditory component of AV targets (A within AV) or a second sound presented at varying delays (1-20 s) after AV exposure (A2 after AV). AV targets were trains of brief presentations (1 or 20), covering a ±30° azimuthal range, and with ±8° (R or L) disparity. We found that the magnitude of visual capture generally reached its peak within a single AV pair and did not dissipate with time, while the ventriloquism aftereffect accumulated with repetitions of AV pairs and dissipated with time. Additionally, the magnitude of the auditory shift induced by each phenomenon was uncorrelated across listeners and visual capture was unaffected by subsequent auditory targets, indicating that visual capture and the ventriloquism aftereffect are separate mechanisms with distinct effects on auditory spatial perception. Our results indicate that visual capture is a 'sample-and-hold' process that binds related objects and stores the combined percept in memory, whereas the ventriloquism aftereffect is a 'leaky integrator' process that accumulates with experience and decays with time to compensate for cross-modal disparities.

Bayesian Alternation during Tactile Augmentation

A large number of studies suggest that the integration of multisensory signals by humans is well-described by Bayesian principles. However, there are very few reports about cue combination between a native and an augmented sense. In particular, we asked the question whether adult participants are able to integrate an augmented sensory cue with existing native sensory information. Hence for the purpose of this study, we build a tactile augmentation device. Consequently, we compared different hypotheses of how untrained adult participants combine information from a native and an augmented sense. In a two-interval forced choice (2 IFC) task, while subjects were blindfolded and seated on a rotating platform, our sensory augmentation device translated information on whole body yaw rotation to tactile stimulation. Three conditions were realized: tactile stimulation only (augmented condition), rotation only (native condition), and both augmented and native information (bimodal condition). Participants had to choose one out of two consecutive rotations with higher angular rotation. For the analysis, we fitted the participants' responses with a probit model and calculated the just notable difference (JND). Then, we compared several models for predicting bimodal from unimodal responses. An objective Bayesian alternation model yielded a better prediction (χred2 = 1.67) than the Bayesian integration model (χred2 = 4.34). Slightly higher accuracy showed a non-Bayesian winner takes all (WTA) model (χred2 = 1.64), which either used only native or only augmented values per subject for prediction. However, the performance of the Bayesian alternation model could be substantially improved (χred2 = 1.09) utilizing subjective weights obtained by a questionnaire. As a result, the subjective Bayesian alternation model predicted bimodal performance most accurately among all tested models. These results suggest that information from augmented and existing sensory modalities in untrained humans is combined via a subjective Bayesian alternation process. Therefore, we conclude that behavior in our bimodal condition is explained better by top down-subjective weighting than by bottom-up weighting based upon objective cue reliability.

Uncertainty leads to persistent effects on reach representations in dorsal premotor cortex

Every movement we make represents one of many possible actions. In reaching tasks with multiple targets, dorsal premotor cortex (PMd) appears to represent all possible actions simultaneously. However, in many situations we are not presented with explicit choices. Instead, we must estimate the best action based on noisy information and execute it while still uncertain of our choice. Here we asked how both primary motor cortex (M1) and PMd represented reach direction during a task in which a monkey made reaches based on noisy, uncertain target information. We found that with increased uncertainty, neurons in PMd actually enhanced their representation of unlikely movements throughout both planning and execution. The magnitude of this effect was highly variable across sessions, and was correlated with a measure of the monkeys’ behavioral uncertainty. These effects were not present in M1. Our findings suggest that PMd represents and maintains a full distribution of potentially correct actions.

DOI:http://dx.doi.org/10.7554/eLife.14316.001

Whether it is trying to find the light switch in a dimly lit room or reaching for your glasses when you wake in the morning, we often need to reach toward objects that we cannot see clearly. In these situations, we plan our movements based both on the limited sensory information that is available, as well as what we have learned from similar situations in the past.

The brain areas involved in using information to decide on the best movement plan appear to be different from those involved in actually executing that plan. One area in particular, called the dorsal premotor cortex (or PMd), is thought to help a person decide where to reach when they are presented with two or more alternative targets. However, it was not known how this brain area is involved in choosing a direction to reach when the targets are fuzzy, or unable to be seen clearly.

Dekleva et al. trained Rhesus macaque monkeys to reach in various directions, towards targets that were represented by fuzzy, uncertain visual cues. These targets were not simply positioned randomly; instead they were more likely to require reaches in certain directions over other directions. Because there were many such training and experimental sessions, the monkeys were able to learn where targets were more likely to be located. Dekleva et al. found that, like humans, the monkeys combined this knowledge from previous experience with the fuzzy visual information; like people, the monkeys also weighted each source of information based on how well they trusted it. For example, blurrier targets were treated as less trustworthy.

Further analysis showed that neurons in the PMd signaled the chosen direction well before the monkey began to reach. However, throughout the entire time the monkey was reaching, the same neurons also seemed to hold in reserve the other, less likely reach directions. In contrast, neurons in the area of the brain that directly controls movement – the primary motor cortex – only ever signaled the direction in which the monkey actually reached.

Further work is now needed to understand the decision-making process that appears to start in the PMd and resolve in the primary motor cortex. In particular, future experiments could explore why the retained information about other possible reach decisions persists throughout the movement, including if this helps the individual to rapidly correct errors or to slowly improve movements over time.

DOI:http://dx.doi.org/10.7554/eLife.14316.002

Intra-Auditory Integration Improves Motor Performance and Synergy in an Accurate Multi-Finger Pressing Task

Humans detect changes in the air pressure and understand the surroundings through the auditory system. The sound humans perceive is composed of two distinct physical properties, frequency and intensity. However, our knowledge is limited how the brain perceives and combines these two properties simultaneously (i.e., intra-auditory integration), especially in relation to motor behaviors. Here, we investigated the effect of intra-auditory integration between the frequency and intensity components of auditory feedback on motor outputs in a constant finger-force production task. The hierarchical variability decomposition model previously developed was used to decompose motor performance into mathematically independent components each of which quantifies a distinct motor behavior such as consistency, repeatability, systematic error, within-trial synergy, or between-trial synergy. We hypothesized that feedback on two components of sound as a function of motor performance (frequency and intensity) would improve motor performance and multi-finger synergy compared to feedback on just one component (frequency or intensity). Subjects were instructed to match the reference force of 18 N with the sum of all finger forces (virtual finger or VF force) while listening to auditory feedback of their accuracy. Three experimental conditions were used: (i) condition F, where frequency changed; (ii) condition I, where intensity changed; (iii) condition FI, where both frequency and intensity changed. Motor performance was enhanced for the FI conditions as compared to either the F or I condition alone. The enhancement of motor performance was achieved mainly by the improved consistency and repeatability. However, the systematic error remained unchanged across conditions. Within- and between-trial synergies were also improved for the FI condition as compared to either the F or I condition alone. However, variability of individual finger forces for the FI condition was not significantly decreased as compared to I condition alone. This result indicates an improvement in motor performance is consistent with Bayesian estimation, and changes in multi-finger interaction mostly result in the enhanced motor performance. These findings provide evidence that the central nervous system can take advantage of the intra-auditory integration in a statistically optimal (Bayesian) fashion to enhance motor performance by improving multi-finger synergy.

Visual capture of a stereo sound: Interactions between cue reliability, sound localization variability, and cross-modal bias

Multisensory interactions involve coordination and sometimes competition between multiple senses. Vision usually dominates audition in spatial judgments when light and sound stimuli are presented from two different physical locations. This study investigated the influence of vision on the perceived location of a phantom sound source placed in a stereo sound field using a pair of loudspeakers emitting identical signals that were delayed or attenuated relative to each other. Results show that although a similar horizontal range (+/-45°) was reported for timing-modulated and level-modulated signals, listeners' localization performance showed greater variability for the timing signals. When visual stimuli were presented simultaneously with the auditory stimuli, listeners showed stronger visual bias for timing-modulated signals than level-modulated and single-speaker control signals. Trial-to-trial errors remained relatively stable over time, suggesting that sound localization uncertainty has an immediate and long-lasting effect on the across-modal bias. Binaural signal analyses further reveal that interaural differences of time and intensity-the two primary cues for sound localization in the azimuthal plane-are inherently more ambiguous for signals placed using timing. These results suggest that binaural ambiguity is intrinsically linked with localization variability and the strength of cross-modal bias in sound localization.

A Model of the Neural Mechanisms Underlying Multisensory Integration in the Superior Colliculus

Much of the information about multisensory integration is derived from studies of the cat superior colliculus (SC), a midbrain structure involved in orientation behaviors. This integration is apparent in the enhanced responses of SC neurons to cross-modal stimuli, responses that exceed those to any of the modality-specific component stimuli. The simplest model of multisensory integration is one in which the SC neuron simply sums its various sensory inputs. However, a number of empirical findings reveal the inadequacy of such a model; for example, the finding that deactivation of cortico-collicular inputs eliminates the enhanced response to a cross-modal stimulus without eliminating responses to the modality-specific component stimuli. These and other empirical findings inform a computational model that accounts for all of the most fundamental aspects of SC multisensory integration. The model is presented in two forms: an algebraic form that conveys the essential insights, and a compartmental form that represents the neuronal computations in a more biologically realistic way.

Hearing Gestures, Seeing Music: Vision Influences Perceived Tone Duration

Percussionists inadvertently use visual information to strategically manipulate audience perception of note duration. Videos of long (L) and short (S) notes performed by a world-renowned percussionist were separated into visual (Lv, Sv) and auditory (La, Sa) components. Visual components contained only the gesture used to perform the note, auditory components the acoustic note itself. Audio and visual components were then crossed to create realistic musical stimuli. Participants were informed of the mismatch, and asked to rate note duration of these audio-visual pairs based on sound alone. Ratings varied based on visual (Lv versus Sv), but not auditory (La versus Sa) components. Therefore while longer gestures do not make longer notes, longer gestures make longer sounding notes through the integration of sensory information. This finding contradicts previous research showing that audition dominates temporal tasks such as duration judgment.

Interval timing, temporal averaging, and cue integration

In a series of recent experiments, we found that if rats are presented with two temporal cues, each signifying that reward will be delivered after a different duration elapses (e.g., tone-10 seconds / light-20 seconds), they will behave as if they have computed a weighted average of these respective durations. In the current article, we argue that this effect, referred to as "temporal averaging", can be understood within the context of Bayesian Decision Theory. Specifically, we propose and provide preliminary data showing that, when averaging, rats weight different durations based on the relative variability of the information their respective cues provide.

Visual Presentation Effects on Identification of Multiple Environmental Sounds

This study examined how the contents and timing of a visual stimulus affect the identification of mixed sounds recorded in a daily life environment. For experiments, we presented four environment sounds as auditory stimuli for 5 s along with a picture or a written word as a visual stimulus that might or might not denote the source of one of the four sounds. Three conditions of temporal relations between the visual stimuli and sounds were used. The visual stimulus was presented either: (a) for 5 s simultaneously with the sound; (b) for 5 s, 1 s before the sound (SOA between the audio and visual stimuli was 6 s); or (c) for 33 ms, 1 s before the sound (SOA was 1033 ms). Participants reported all identifiable sounds for those audio-visual stimuli. To characterize the effects of visual stimuli on sound identification, the following were used: the identification rates of sounds for which the visual stimulus denoted its sound source, the rates of other sounds for which the visual stimulus did not denote the sound source, and the frequency of false hearing of a sound that was not presented for each sound set. Results of the four experiments demonstrated that a picture or a written word promoted identification of the sound when it was related to the sound, particularly when the visual stimulus was presented for 5 s simultaneously with the sounds. However, a visual stimulus preceding the sounds had a benefit only for the picture, not for the written word. Furthermore, presentation with a picture denoting a sound simultaneously with the sound reduced the frequency of false hearing. These results suggest three ways that presenting a visual stimulus affects identification of the auditory stimulus. First, activation of the visual representation extracted directly from the picture promotes identification of the denoted sound and suppresses the processing of sounds for which the visual stimulus did not denote the sound source. Second, effects based on processing of the conceptual information promote identification of the denoted sound and suppress the processing of sounds for which the visual stimulus did not denote the sound source. Third, processing of the concurrent visual representation suppresses false hearing.

Ventriloquism effect with sound stimuli varying in both azimuth and elevation

When presented with a spatially discordant auditory-visual stimulus, subjects sometimes perceive the sound and the visual stimuli as coming from the same location. Such a phenomenon is often referred to as perceptual fusion or ventriloquism, as it evokes the illusion created by a ventriloquist when his voice seems to emanate from his puppet rather than from his mouth. While this effect has been extensively examined in the horizontal plane and to a lesser extent in distance, few psychoacoustic studies have focused on elevation. In the present experiment, sequences of a man talking were presented to subjects. His voice could be reproduced on different loudspeakers, which created disparities in both azimuth and elevation between the sound and the visual stimuli. For each presentation, subjects had to indicate whether the voice seemed to emanate from the mouth of the actor or not. Results showed that ventriloquism could be observed with larger audiovisual disparities in elevation than in azimuth.

Biases in Visual, Auditory, and Audiovisual Perception of Space

Localization of objects and events in the environment is critical for survival, as many perceptual and motor tasks rely on estimation of spatial location. Therefore, it seems reasonable to assume that spatial localizations should generally be accurate. Curiously, some previous studies have reported biases in visual and auditory localizations, but these studies have used small sample sizes and the results have been mixed. Therefore, it is not clear (1) if the reported biases in localization responses are real (or due to outliers, sampling bias, or other factors), and (2) whether these putative biases reflect a bias in sensory representations of space or a priori expectations (which may be due to the experimental setup, instructions, or distribution of stimuli). Here, to address these questions, a dataset of unprecedented size (obtained from 384 observers) was analyzed to examine presence, direction, and magnitude of sensory biases, and quantitative computational modeling was used to probe the underlying mechanism(s) driving these effects. Data revealed that, on average, observers were biased towards the center when localizing visual stimuli, and biased towards the periphery when localizing auditory stimuli. Moreover, quantitative analysis using a Bayesian Causal Inference framework suggests that while pre-existing spatial biases for central locations exert some influence, biases in the sensory representations of both visual and auditory space are necessary to fully explain the behavioral data. How are these opposing visual and auditory biases reconciled in conditions in which both auditory and visual stimuli are produced by a single event? Potentially, the bias in one modality could dominate, or the biases could interact/cancel out. The data revealed that when integration occurred in these conditions, the visual bias dominated, but the magnitude of this bias was reduced compared to unisensory conditions. Therefore, multisensory integration not only improves the precision of perceptual estimates, but also the accuracy.

Temporal averaging across multiple response options: insight into the mechanisms underlying integration

Rats trained on a dual-duration, dual-modality peak-interval procedure (e.g., tone = 10 s/light = 20 s) often show unimodal response distributions with peaks that fall in between the anchor durations when both cues are presented as a simultaneous compound. Two hypotheses can explain this finding. According to the averaging hypothesis, rats integrate the anchor durations into an average during compound trials, with each duration being weighted by its respective reinforcement probability. According to the simultaneous temporal processing hypothesis, rats time both durations veridically and simultaneously during compound trials and respond continuously across both durations, thereby producing a unimodal response distribution with a peak falling in between the anchor durations. In the present compounding experiment, rats were trained to associate a tone and light with two different durations (e.g., 5 and 20 s, respectively). However, in contrast to previous experiments, each cue was also associated with a distinct response requirement (e.g., left nosepoke for tone/right nosepoke for light). On the majority of compound trials, responding on a given nosepoke fell close to its respective duration, but was shifted in the direction of the other cue's duration, suggesting rats timed an average of the two durations. However, more weight appeared to be given to the duration associated with the manipulandum on which the rat responded, rather than the duration associated with a higher reinforcement probability as predicted by the averaging hypothesis. Group differences were also observed, with rats trained to associate the tone and light with the short and long durations, respectively, being more likely to show these shifts than the counterbalanced modality-duration group (i.e., light-short/tone-long). This parallels group differences observed in past studies and suggest that cue weighting in response to stimulus compounds is influenced by the modality-duration relationship of the anchor cues. The current results suggest that temporal averaging is a more flexible process than previously theorized and provide novel insight into the mechanisms that affect cue weighting.

The interaction of vision and audition in two-dimensional space

Using a mouse-driven visual pointer, 10 participants made repeated open-loop egocentric localizations of memorized visual, auditory, and combined visual-auditory targets projected randomly across the two-dimensional frontal field (2D). The results are reported in terms of variable error, constant error and local distortion. The results confirmed that auditory and visual maps of the egocentric space differ in their precision (variable error) and accuracy (constant error), both from one another and as a function of eccentricity and direction within a given modality. These differences were used, in turn, to make predictions about the precision and accuracy within which spatially and temporally congruent bimodal visual-auditory targets are localized. Overall, the improvement in precision for bimodal relative to the best unimodal target revealed the presence of optimal integration well-predicted by the Maximum Likelihood Estimation (MLE) model. Conversely, the hypothesis that accuracy in localizing the bimodal visual-auditory targets would represent a compromise between auditory and visual performance in favor of the most precise modality was rejected. Instead, the bimodal accuracy was found to be equivalent to or to exceed that of the best unimodal condition. Finally, we described how the different types of errors could be used to identify properties of the internal representations and coordinate transformations within the central nervous system (CNS). The results provide some insight into the structure of the underlying sensorimotor processes employed by the brain and confirm the usefulness of capitalizing on naturally occurring differences between vision and audition to better understand their interaction and their contribution to multimodal perception.

Duration estimates within a modality are integrated sub-optimally

Perceived duration can be influenced by various properties of sensory stimuli. For example, visual stimuli of higher temporal frequency are perceived to last longer than those of lower temporal frequency. How does the brain form a representation of duration when each of two simultaneously presented stimuli influences perceived duration in different way? To answer this question, we investigated the perceived duration of a pair of dynamic visual stimuli of different temporal frequencies in comparison to that of a single visual stimulus of either low or high temporal frequency. We found that the duration representation of simultaneously occurring visual stimuli is best described by weighting the estimates of duration based on each individual stimulus. However, the weighting performance deviates from the prediction of statistically optimal integration. In addition, we provided a Bayesian account to explain a difference in the apparent sensitivity of the psychometric curves introduced by the order in which the two stimuli are displayed in a two-alternative forced-choice task.

Crossmodal integration improves sensory detection thresholds in the ferret

During the last two decades ferrets (Mustela putorius) have been established as a highly efficient animal model in different fields in neuroscience. Here we asked whether ferrets integrate sensory information according to the same principles established for other species. Since only few methods and protocols are available for behaving ferrets we developed a head-free, body-restrained approach allowing a standardized stimulation position and the utilization of the ferret’s natural response behavior. We established a behavioral paradigm to test audiovisual integration in the ferret. Animals had to detect a brief auditory and/or visual stimulus presented either left or right from their midline. We first determined detection thresholds for auditory amplitude and visual contrast. In a second step, we combined both modalities and compared psychometric fits and the reaction times between all conditions. We employed Maximum Likelihood Estimation (MLE) to model bimodal psychometric curves and to investigate whether ferrets integrate modalities in an optimal manner. Furthermore, to test for a redundant signal effect we pooled the reaction times of all animals to calculate a race model. We observed that bimodal detection thresholds were reduced and reaction times were faster in the bimodal compared to unimodal conditions. The race model and MLE modeling showed that ferrets integrate modalities in a statistically optimal fashion. Taken together, the data indicate that principles of multisensory integration previously demonstrated in other species also apply to crossmodal processing in the ferret.

Allocentric coding: spatial range and combination rules

When a visual target is presented with neighboring landmarks, its location can be determined both relative to the self (egocentric coding) and relative to these landmarks (allocentric coding). In the present study, we investigated (1) how allocentric coding depends on the distance between the targets and their surrounding landmarks (i.e. the spatial range) and (2) how allocentric and egocentric coding interact with each other across targets-landmarks distances (i.e. the combination rules). Subjects performed a memory-based pointing task toward previously gazed targets briefly superimposed (200ms) on background images of cluttered city landscapes. A variable portion of the images was occluded in order to control the distance between the targets and the closest potential landmarks within those images. The pointing responses were performed after large saccades and the reappearance of the images at their initial location. However, in some trials, the images' elements were slightly shifted (±3°) in order to introduce a subliminal conflict between the allocentric and egocentric reference frames. The influence of allocentric coding in the pointing responses was found to decrease with increasing target-landmarks distances, although it remained significant even at the largest distances (⩾10°). Interestingly, both the decreasing influence of allocentric coding and the concomitant increase in pointing responses variability were well captured by a Bayesian model in which the weighted combination of allocentric and egocentric cues is governed by a coupling prior.

Learning multisensory representations for auditory-visual transfer of sequence category knowledge: a probabilistic language of thought approach

If a person is trained to recognize or categorize objects or events using one sensory modality, the person can often recognize or categorize those same (or similar) objects and events via a novel modality. This phenomenon is an instance of cross-modal transfer of knowledge. Here, we study the Multisensory Hypothesis which states that people extract the intrinsic, modality-independent properties of objects and events, and represent these properties in multisensory representations. These representations underlie cross-modal transfer of knowledge. We conducted an experiment evaluating whether people transfer sequence category knowledge across auditory and visual domains. Our experimental data clearly indicate that we do. We also developed a computational model accounting for our experimental results. Consistent with the probabilistic language of thought approach to cognitive modeling, our model formalizes multisensory representations as symbolic "computer programs" and uses Bayesian inference to learn these representations. Because the model demonstrates how the acquisition and use of amodal, multisensory representations can underlie cross-modal transfer of knowledge, and because the model accounts for subjects' experimental performances, our work lends credence to the Multisensory Hypothesis. Overall, our work suggests that people automatically extract and represent objects' and events' intrinsic properties, and use these properties to process and understand the same (and similar) objects and events when they are perceived through novel sensory modalities.

Effects of virtual speaker density and room reverberation on spatiotemporal thresholds of audio-visual motion coherence

The present study examined the effects of spatial sound-source density and reverberation on the spatiotemporal window for audio-visual motion coherence. Three different acoustic stimuli were generated in Virtual Auditory Space: two acoustically "dry" stimuli via the measurement of anechoic head-related impulse responses recorded at either 1° or 5° spatial intervals (Experiment 1), and a reverberant stimulus rendered from binaural room impulse responses recorded at 5° intervals in situ in order to capture reverberant acoustics in addition to head-related cues (Experiment 2). A moving visual stimulus with invariant localization cues was generated by sequentially activating LED's along the same radial path as the virtual auditory motion. Stimuli were presented at 25°/s, 50°/s and 100°/s with a random spatial offset between audition and vision. In a 2AFC task, subjects made a judgment of the leading modality (auditory or visual). No significant differences were observed in the spatial threshold based on the point of subjective equivalence (PSE) or the slope of psychometric functions (β) across all three acoustic conditions. Additionally, both the PSE and β did not significantly differ across velocity, suggesting a fixed spatial window of audio-visual separation. Findings suggest that there was no loss in spatial information accompanying the reduction in spatial cues and reverberation levels tested, and establish a perceptual measure for assessing the veracity of motion generated from discrete locations and in echoic environments.

Reward maximization justifies the transition from sensory selection at childhood to sensory integration at adulthood

In a multisensory task, human adults integrate information from different sensory modalities -behaviorally in an optimal Bayesian fashion- while children mostly rely on a single sensor modality for decision making. The reason behind this change of behavior over age and the process behind learning the required statistics for optimal integration are still unclear and have not been justified by the conventional Bayesian modeling. We propose an interactive multisensory learning framework without making any prior assumptions about the sensory models. In this framework, learning in every modality and in their joint space is done in parallel using a single-step reinforcement learning method. A simple statistical test on confidence intervals on the mean of reward distributions is used to select the most informative source of information among the individual modalities and the joint space. Analyses of the method and the simulation results on a multimodal localization task show that the learning system autonomously starts with sensory selection and gradually switches to sensory integration. This is because, relying more on modalities -i.e. selection- at early learning steps (childhood) is more rewarding than favoring decisions learned in the joint space since, smaller state-space in modalities results in faster learning in every individual modality. In contrast, after gaining sufficient experiences (adulthood), the quality of learning in the joint space matures while learning in modalities suffers from insufficient accuracy due to perceptual aliasing. It results in tighter confidence interval for the joint space and consequently causes a smooth shift from selection to integration. It suggests that sensory selection and integration are emergent behavior and both are outputs of a single reward maximization process; i.e. the transition is not a preprogrammed phenomenon.

Optimal multisensory decision-making in a reaction-time task

Humans and animals can integrate sensory evidence from various sources to make decisions in a statistically near-optimal manner, provided that the stimulus presentation time is fixed across trials. Little is known about whether optimality is preserved when subjects can choose when to make a decision (reaction-time task), nor when sensory inputs have time-varying reliability. Using a reaction-time version of a visual/vestibular heading discrimination task, we show that behavior is clearly sub-optimal when quantified with traditional optimality metrics that ignore reaction times. We created a computational model that accumulates evidence optimally across both cues and time, and trades off accuracy with decision speed. This model quantitatively explains subjects's choices and reaction times, supporting the hypothesis that subjects do, in fact, accumulate evidence optimally over time and across sensory modalities, even when the reaction time is under the subject's control.

Feature-positive discriminations during a spatial-search task with humans

During feature-positive operant discriminations, a conditional cue, X, signals whether responses made during a second stimulus, A, are reinforced. Few studies have examined how landmarks, which can be trained to control the spatial distribution of responses during search tasks, might operate under conditional control. We trained college students to search for a target hidden on a computer monitor. Participants learned that responses to a hidden target location signaled by a landmark (e.g., A) would be reinforced only if the landmark was preceded by a colored background display (e.g., X). In Experiment 1, participants received feature-positive training (+←YB/ XA→+/A-/B-) with the hidden target to the right of A and to left of B. Responding during nonreinforced transfer test trials (XB-/YA-) indicated conditional control by the colored background, and spatial accuracy indicated a greater weighting of spatial information provided by the landmark than by the conditional cue. In Experiments 2a and 2b, the location of the target relative to landmark A was conditional on the colored background (+←YA/ XA→+/ ZB→+/ +←C /A-/B-). At test, conditional control and a greater weighting for the landmark's spatial information were again found, but we also report evidence for spatial interference by the conditional stimulus. Overall, we found that hierarchical accounts best explain the observed differences in response magnitude, whereas spatial accuracy was best explained via spatial learning models that emphasize the reliability, stability, and proximity of landmarks to a target.

Intersensory binding across space and time: a tutorial review

Spatial ventriloquism refers to the phenomenon that a visual stimulus such as a flash can attract the perceived location of a spatially discordant but temporally synchronous sound. An analogous example of mutual attraction between audition and vision has been found in the temporal domain, where temporal aspects of a visual event, such as its onset, frequency, or duration, can be biased by a slightly asynchronous sound. In this review, we examine various manifestations of spatial and temporal attraction between the senses (both direct effects and aftereffects), and we discuss important constraints on the occurrence of these effects. Factors that potentially modulate ventriloquism-such as attention, synesthetic correspondence, and other cognitive factors-are described. We trace theories and models of spatial and temporal ventriloquism, from the traditional unity assumption and modality appropriateness hypothesis to more recent Bayesian and neural network approaches. Finally, we summarize recent evidence probing the underlying neural mechanisms of spatial and temporal ventriloquism.

Relation between belief and performance in perceptual decision making

In an uncertain and ambiguous world, effective decision making requires that subjects form and maintain a belief about the correctness of their choices, a process called meta-cognition. Prediction of future outcomes and self-monitoring are only effective if belief closely matches behavioral performance. Equality between belief and performance is also critical for experimentalists to gain insight into the subjects' belief by simply measuring their performance. Assuming that the decision maker holds the correct model of the world, one might indeed expect that belief and performance should go hand in hand. Unfortunately, we show here that this is rarely the case when performance is defined as the percentage of correct responses for a fixed stimulus, a standard definition in psychophysics. In this case, belief equals performance only for a very narrow family of tasks, whereas in others they will only be very weakly correlated. As we will see it is possible to restore this equality in specific circumstances but this remedy is only effective for a decision-maker, not for an experimenter. We furthermore show that belief and performance do not match when conditioned on task difficulty--as is common practice when plotting the psychometric curve--highlighting common pitfalls in previous neuroscience work. Finally, we demonstrate that miscalibration and the hard-easy effect observed in humans' and other animals' certainty judgments could be explained by a mismatch between the experimenter's and decision maker's expected distribution of task difficulties. These results have important implications for experimental design and are of relevance for theories that aim to unravel the nature of meta-cognition.

Effects of augmentative visual training on audio-motor mapping

The purpose of this study was to determine the effect of augmentative visual feedback training on auditory-motor performance. Thirty-two healthy young participants used facial surface electromyography (sEMG) to control a human-machine interface (HMI) for which the output was vowel synthesis. An auditory-only (AO) group (n=16) trained with auditory feedback alone and an auditory-visual (AV) group (n=16) trained with auditory feedback and progressively-removed visual feedback. Subjects participated in three training sessions and one testing session over 3days. During the testing session they were given novel targets to test auditory-motor generalization. We hypothesized that the auditory-visual group would perform better on the novel set of targets than the group that trained with auditory feedback only. Analysis of variance on the percentage of total targets reached indicated a significant interaction between group and session: individuals in the AV group performed significantly better than those in the AO group during early training sessions (while using visual feedback), but no difference was seen between the two groups during later sessions. Results suggest that augmentative visual feedback during training does not improve auditory-motor performance.

Bridging the gap between theories of sensory cue integration and the physiology of multisensory neurons

The richness of perceptual experience, as well as its usefulness for guiding behaviour, depends on the synthesis of information across multiple senses. Recent decades have witnessed a surge in our understanding of how the brain combines sensory cues. Much of this research has been guided by one of two distinct approaches: one is driven primarily by neurophysiological observations, and the other is guided by principles of mathematical psychology and psychophysics. Conflicting results and interpretations have contributed to a conceptual gap between psychophysical and physiological accounts of cue integration, but recent studies of visual-vestibular cue integration have narrowed this gap considerably.

Outcome dependency alters the neural substrates of impression formation

How do people maintain consistent impressions of other people when other people are often inconsistent? The present research addresses this question by combining recent neuroscientific insights with ecologically meaningful behavioral methods. Participants formed impressions of real people whom they met in a personally involving situation. fMRI and supporting behavioral data revealed that outcome dependency (i.e., depending on another person for a desired outcome) alters previously identified neural dynamics of impression formation. Consistent with past research, a functional localizer identified a region of dorsomedial PFC previously linked to social impression formation. In the main task, this ROI revealed the predicted patterns of activity across outcome dependency conditions: greater BOLD response when information confirmed (vs. violated) social expectations if participants were outcome-independent, and the reverse pattern if participants were outcome-dependent. We suggest that, although social perceivers often discount expectancy-disconfirming information as noise, being dependent on another person for a desired outcome focuses impression-formation processing on the most diagnostic information, rather than on the most tractable information.

Audio-Visual Temporal Recalibration Can be Constrained by Content Cues Regardless of Spatial Overlap

It has now been well established that the point of subjective synchrony for audio and visual events can be shifted following exposure to asynchronous audio-visual presentations, an effect often referred to as temporal recalibration. Recently it was further demonstrated that it is possible to concurrently maintain two such recalibrated estimates of audio-visual temporal synchrony. However, it remains unclear precisely what defines a given audio-visual pair such that it is possible to maintain a temporal relationship distinct from other pairs. It has been suggested that spatial separation of the different audio-visual pairs is necessary to achieve multiple distinct audio-visual synchrony estimates. Here we investigated if this is necessarily true. Specifically, we examined whether it is possible to obtain two distinct temporal recalibrations for stimuli that differed only in featural content. Using both complex (audio visual speech; see Experiment 1) and simple stimuli (high and low pitch audio matched with either vertically or horizontally oriented Gabors; see Experiment 2) we found concurrent, and opposite, recalibrations despite there being no spatial difference in presentation location at any point throughout the experiment. This result supports the notion that the content of an audio-visual pair alone can be used to constrain distinct audio-visual synchrony estimates regardless of spatial overlap.

Bayesian prediction and evaluation in the anterior cingulate cortex

The dorsal anterior cingulate cortex (dACC) has been implicated in a variety of cognitive control functions, among them the monitoring of conflict, error, and volatility, error anticipation, reward learning, and reward prediction errors. In this work, we used a Bayesian ideal observer model, which predicts trial-by-trial probabilistic expectation of stop trials and response errors in the stop-signal task, to differentiate these proposed functions quantitatively. We found that dACC hemodynamic response, as measured by functional magnetic resonance imaging, encodes both the absolute prediction error between stimulus expectation and outcome, and the signed prediction error related to response outcome. After accounting for these factors, dACC has no residual correlation with conflict or error likelihood in the stop-signal task. Consistent with recent monkey neural recording studies, and in contrast with other neuroimaging studies, our work demonstrates that dACC reports at least two different types of prediction errors, and beyond contexts that are limited to reward processing.

When more is less: increasing allocentric visual information can switch visual-proprioceptive combination from an optimal to sub-optimal process

When reaching for an object in the environment, the brain often has access to multiple independent estimates of that object's location. For example, if someone places their coffee cup on a table, then later they know where it is because they see it, but also because they remember how their reaching limb was oriented when they placed the cup. Intuitively, one would expect more accurate reaches if either of these estimates were improved (e.g., if a light were turned on so the cup were more visible). It is now well-established that the brain tends to combine two or more estimates about the same stimulus as a maximum-likelihood estimator (MLE), which is the best thing to do when estimates are unbiased. Even in the presence of small biases, relying on the MLE rule is still often better than choosing a single estimate. For this work, we designed a reaching task in which human subjects could integrate proprioceptive and allocentric (landmark-relative) visual information to reach for a remembered target. Even though both of these modalities contain some level of bias, we demonstrate via simulation that our subjects should use an MLE rule in preference to relying on one modality or the other in isolation. Furthermore, we show that when visual information is poor, subjects do, indeed, combine information in this way. However, when we improve the quality of visual information, subjects counter-intuitively switch to a sub-optimal strategy that occasionally includes reliance on a single modality.

Grey matter volume in early human visual cortex predicts proneness to the sound-induced flash illusion

Visual perception can be modulated by sounds. A drastic example of this is the sound-induced flash illusion: when a single flash is accompanied by two bleeps, it is sometimes perceived in an illusory fashion as two consecutive flashes. However, there are strong individual differences in proneness to this illusion. Some participants experience the illusion on almost every trial, whereas others almost never do. We investigated whether such individual differences in proneness to the sound-induced flash illusion were reflected in structural differences in brain regions whose activity is modulated by the illusion. We found that individual differences in proneness to the illusion were strongly and significantly correlated with local grey matter volume in early retinotopic visual cortex. Participants with smaller early visual cortices were more prone to the illusion. We propose that strength of auditory influences on visual perception is determined by individual differences in recurrent connections, cross-modal attention and/or optimal weighting of sensory channels.

Bayesian Models of Brain and Behaviour

This paper presents a review of Bayesian models of brain and behaviour. We first review the basic principles of Bayesian inference. This is followed by descriptions of sampling and variational methods for approximate inference, and forward and backward recursions in time for inference in dynamical models. The review of behavioural models covers work in visual processing, sensory integration, sensorimotor integration, and collective decision making. The review of brain models covers a range of spatial scales from synapses to neurons and population codes, but with an emphasis on models of cortical hierarchies. We describe a simple hierarchical model which provides a mathematical framework relating constructs in Bayesian inference to those in neural computation. We close by reviewing recent theoretical developments in Bayesian inference for planning and control.

Self-motion leads to mandatory cue fusion across sensory modalities

When perceiving properties of the world, we effortlessly combine multiple sensory cues into optimal estimates. Estimates derived from the individual cues are generally retained once the multisensory estimate is produced and discarded only if the cues stem from the same sensory modality (i.e., mandatory fusion). Does multisensory integration differ in that respect when the object of perception is one's own body, rather than an external variable? We quantified how humans combine visual and vestibular information for perceiving own-body rotations and specifically tested whether such idiothetic cues are subjected to mandatory fusion. Participants made extensive size comparisons between successive whole body rotations using only visual, only vestibular, and both senses together. Probabilistic descriptions of the subjects' perceptual estimates were compared with a Bayes-optimal integration model. Similarity between model predictions and experimental data echoed a statistically optimal mechanism of multisensory integration. Most importantly, size discrimination data for rotations composed of both stimuli was best accounted for by a model in which only the bimodal estimator is accessible for perceptual judgments as opposed to an independent or additive use of all three estimators (visual, vestibular, and bimodal). Indeed, subjects' thresholds for detecting two multisensory rotations as different from one another were, in pertinent cases, larger than those measured using either single-cue estimate alone. Rotations different in terms of the individual visual and vestibular inputs but quasi-identical in terms of the integrated bimodal estimate became perceptual metamers. This reveals an exceptional case of mandatory fusion of cues stemming from two different sensory modalities.

A common network architecture efficiently implements a variety of sparsity-based inference problems

The sparse coding hypothesis has generated significant interest in the computational and theoretical neuroscience communities, but there remain open questions about the exact quantitative form of the sparsity penalty and the implementation of such a coding rule in neurally plausible architectures. The main contribution of this work is to show that a wide variety of sparsity-based probabilistic inference problems proposed in the signal processing and statistics literatures can be implemented exactly in the common network architecture known as the locally competitive algorithm (LCA). Among the cost functions we examine are approximate l(p) norms (0 ≤ p ≤ 2), modified l(p)-norms, block-l1 norms, and reweighted algorithms. Of particular interest is that we show significantly increased performance in reweighted l1 algorithms by inferring all parameters jointly in a dynamical system rather than using an iterative approach native to digital computational architectures.

Organizing probabilistic models of perception

Probability has played a central role in models of perception for more than a century, but a look at probabilistic concepts in the literature raises many questions. Is being Bayesian the same as being optimal? Are recent Bayesian models fundamentally different from classic signal detection theory models? Do findings of near-optimal inference provide evidence that neurons compute with probability distributions? This review aims to disentangle these concepts and to classify empirical evidence accordingly.

Combining symbolic cues with sensory input and prior experience in an iterative bayesian framework

Perception and action are the result of an integration of various sources of information, such as current sensory input, prior experience, or the context in which a stimulus occurs. Often, the interpretation is not trivial hence needs to be learned from the co-occurrence of stimuli. Yet, how do we combine such diverse information to guide our action? Here we use a distance production-reproduction task to investigate the influence of auxiliary, symbolic cues, sensory input, and prior experience on human performance under three different conditions that vary in the information provided. Our results indicate that subjects can (1) learn the mapping of a verbal, symbolic cue onto the stimulus dimension and (2) integrate symbolic information and prior experience into their estimate of displacements. The behavioral results are explained by to two distinct generative models that represent different structural approaches of how a Bayesian observer would combine prior experience, sensory input, and symbolic cue information into a single estimate of displacement. The first model interprets the symbolic cue in the context of categorization, assuming that it reflects information about a distinct underlying stimulus range (categorical model). The second model applies a multi-modal integration approach and treats the symbolic cue as additional sensory input to the system, which is combined with the current sensory measurement and the subjects' prior experience (cue-combination model). Notably, both models account equally well for the observed behavior despite their different structural assumptions. The present work thus provides evidence that humans can interpret abstract symbolic information and combine it with other types of information such as sensory input and prior experience. The similar explanatory power of the two models further suggest that issues such as categorization and cue-combination could be explained by alternative probabilistic approaches.

Emotion and decision-making: affect-driven belief systems in anxiety and depression

Emotion processing and decision-making are integral aspects of daily life. However, our understanding of the interaction between these constructs is limited. In this review, we summarize theoretical approaches that link emotion and decision-making, and focus on research with anxious or depressed individuals to show how emotions can interfere with decision-making. We integrate the emotional framework based on valence and arousal with a Bayesian approach to decision-making in terms of probability and value processing. We discuss how studies of individuals with emotional dysfunctions provide evidence that alterations of decision-making can be viewed in terms of altered probability and value computation. We argue that the probabilistic representation of belief states in the context of partially observable Markov decision processes provides a useful approach to examine alterations in probability and value representation in individuals with anxiety and depression, and outline the broader implications of this approach.

Evidence for multisensory integration in the elicitation of prior entry by bimodal cues

This study reports an experiment investigating the relative effects of intramodal, crossmodal and bimodal cues on visual and auditory temporal order judgements. Pairs of visual or auditory targets, separated by varying stimulus onset asynchronies, were presented to either side of a central fixation (±45°), and participants were asked to identify the target that had occurred first. In some of the trials, one of the targets was preceded by a short, non-predictive visual, auditory or audiovisual cue stimulus. The cue and target stimuli were presented at the exact same locations in space. The point of subjective simultaneity revealed a consistent spatiotemporal bias towards targets at the cued location. For the visual targets, the intramodal cue elicited the largest, and the crossmodal cue the smallest, bias. The bias elicited by the bimodal cue fell between the intramodal and crossmodal cue biases, with significant differences between all cue types. The pattern for the auditory targets was similar apart from a scaling factor and greater variance, so the differences between the cue conditions did not reach significance. These results provide evidence for multisensory integration in exogenous attentional cueing. The magnitude of the bimodal cueing effect was equivalent to the average of the facilitation elicited by the intramodal and crossmodal cues. Under the assumption that the visual and auditory cues were equally informative, this is consistent with the notion that exogenous attention, like perception, integrates multimodal information in an optimal way.

Knowing how much you don't know: a neural organization of uncertainty estimates

How we estimate uncertainty is important in decision neuroscience and has wide-ranging implications in basic and clinical neuroscience, from computational models of optimality to ideas on psychopathological disorders including anxiety, depression and schizophrenia. Empirical research in neuroscience, which has been based on divergent theoretical assumptions, has focused on the fundamental question of how uncertainty is encoded in the brain and how it influences behaviour. Here, we integrate several theoretical concepts about uncertainty into a decision-making framework. We conclude that the currently available evidence indicates that distinct neural encoding (including summary statistic-type representations) of uncertainty occurs in distinct neural systems.