Individuals with autism show non-adaptive relative weighting of perceptual prior and sensory reliability

Abstract

Modulation in sensory-perceptual processing is a known characteristic of autism, although the underlying mechanism is debated. A prevailing account is formulated in Bayesian terms, where either a reduced prior or reduced noise in the measurement (sensory input) may account for the modulated perception as expressed by the posterior distribution. However, research has shown that individuals with autism use priors in some conditions, and to the same extent as neurotypicals, while other studies fail to show enhanced sensory sensitivity in these individuals. We asked whether the modulated prior effects on perception may arise from non-adaptive relative weighting of priors to sensory reliability. We employed a Two-alternative forced choice (2AFC) width discrimination task, using the width–height illusion, which is based on a long-term acquired bias, where a taller rectangle is typically perceived as thinner than a shorter one. The measurement was manipulated by adding Gaussian blur on the vertical edges of the rectangles. Typically developed individuals displayed the expected increase in bias as a function of noise in the measurement. High-functioning individuals with autism exhibited typical perceptual resolutions and similar susceptibility to the illusion. However, the relative weighting of the perceptual bias and the sensory input differed in their effect on the two groups. Individuals with autism showed a non-adaptive, consistent bias across the different degrees of sensory noise, while typically developed individuals displayed monotonically increasing biases. Cluster analyses showed that this difference in the relative weighting between the groups was preserved regardless of the overall illusion magnitude displayed by individuals in each cluster.

Lay abstract

Unique perceptual skills and abnormalities in perception have been extensively demonstrated in those with autism for many perceptual domains, accounting, at least in part, for some of the main symptoms. Several new hypotheses suggest that perceptual representations in autism are unrefined, appear less constrained by exposure and regularities of the environment, and rely more on actual concrete input. Consistent with these emerging views, a bottom-up, data-driven fashion of processing has been suggested to account for the atypical perception in autism. It is yet unclear, however, whether reduced effects of prior knowledge and top-down information, or rather reduced noise in the sensory input, account for the often-reported bottom-up mode of processing in autism. We show that neither is sufficiently supported. Instead, we demonstrate clear differences between autistics and neurotypicals in how incoming input is weighted against prior knowledge and experience in determining the final percept. Importantly, the findings tap central differences in perception between those with and without autism that are consistent across identified sub-clusters within each group.

Keywords

autism spectrum disorder Bayesian perception inferred perception perceptual illusions width–height illusion

Introduction

Autism spectrum disorders (ASDs) are neurodevelopmental disorders of mostly unknown etiology whose key features include deficits in social communication, repetitive behaviors, restricted areas of interest, and sensory processing difficulties (American Psychiatric Association, 2013). Abnormalities have also been extensively demonstrated in ASD in many perceptual domains, accounting, at least in part, for some of the main symptoms (e.g. Dakin & Frith, 2005). Several new hypotheses suggest that perceptual representations in autism are unrefined and less constrained by exposure, regularities, or psychophysical principles (Hadad et al., 2017, 2019; Hadad & Schwartz, 2019; Van de Cruys et al., 2014), relying more on actual concrete stimuli (Evers et al., 2018).

Consistent with these emerging views, a Bayesian account has posited that a bottom-up, data-driven mode of processing may underlie atypical perception in autism (Pellicano & Burr, 2012). Under this view, perception in autism is thought to rely on the actual incoming input with little or no influence of previous experience and knowledge. In typically developed (TD) individuals, perception has long been described as a process of unconscious inference or automatic “best guesses” about the structure of the world, given the available sensory input and given an internal, biased model of the world (Von Helmholtz, 1867; for a review, see Friston, 2010). This notion has been modeled using Bayesian statistical decision theory, a principled method of reasoning under uncertainty that defines perception as an active process arising from a statistically optimal combination of noisy incoming evidence (i.e. likelihood) and prior knowledge (e.g. Friston, 2010). Internal priors are typically shown to constrain perception from early in life but are reweighted during childhood (Thomas et al., 2010).

It has been strongly claimed that individuals with autism do not seem to optimize perception and behavioral outcomes, showing reduced sensitivity to context, with responses lying closer to the sensory input (Lawson et al., 2014). Some Bayesian models have attributed these modulations to underweighting of prior knowledge because of wider internal priors (less precise contextual representations; Pellicano & Burr, 2012) and others to precise likelihood (Brock, 2012; Lawson et al., 2014). Underweighting of prior knowledge has been attributed to difficulties in formulating the prior (Pellicano & Burr, 2012), or to high and inflexible precision of (sensory) prediction errors (Van de Cruys et al., 2014). However, contrary to these claims of uniformly weak priors (Pellicano & Burr, 2012), there is evidence for intact use of perceptual expectations and prior belief in ASD, for both natural and task-related priors (e.g. Feigin et al., 2021; Hadad & Schwartz, 2019; Manning et al., 2017; Van de Cruys et al., 2021). Several studies report that individuals with autism often develop very strong priors or expectations in particular contexts. This has been shown, for example, in studies measuring event-related potential (ERP) responses to deviant (unexpected) stimuli, demonstrating significant mismatch negativity responses (i.e. reaction to (small) violations of predictions), thus implying the ability of individuals with ASD to form priors or expectations (Ferri et al., 2003; Kujala et al., 2007).

Other accounts of the modulated effects of priors on perception in ASD have focused on how priors develop on the basis of prediction errors. It has been suggested that the high and inflexible precision of prediction errors may more naturally account for strong priors or expectations that individuals with autism may develop in particular contexts but do not generalize (Van de Cruys et al., 2014). Specifically, in simple visual discrimination tasks, individuals with ASD seem able to implicitly learn a prior mean of the presented stimuli but fail to flexibly adjust the prior precision to the context (Sapey-Triomphe et al., 2020). They have also been shown to overestimate the volatility of the sensory environment, at the expense of learning to build stable expectations (Lawson et al., 2017). In many of these studies, however, findings of seemingly lower flexibility in autism when adjusting the prior may actually reflect slower rates of building up a precise representation of the prior. This interpretation is supported by findings of overall slower learning of prior information (Soulières et al., 2011) and slower updating of the prior based on recent stimuli history (Lieder et al., 2019).

Although precise likelihood may also account for weakening of priors (Brock, 2012), many models do not test sensory processing itself and assume it is intact (Teufel et al., 2013). As either a reduced prior or reduced noise in the measurement (sensory input) may account for the altered perception in autism, any claim of modulated effects of the priors should consider the possibility that they arise from alternations in the sensory processing as possibly affecting the likelihood function.

Indeed, it has recently been found that sensory input is registered differently in the sensory-perceptual system in ASD, demonstrating a striking violation of basic psychophysical principles (Hadad & Schwartz, 2019). Consequently, the typical pattern of increased effects of biases for noisier measurements (e.g. Petzschner et al., 2015) may not be evident in autism where scalar variability (i.e. Weber ratio) does not seem to hold.

Here, we examine how an incoming input is weighted against prior knowledge and experience in determining the final percept in autism. We manipulated stimulus noise and tested possible modulations in the balance between measurement reliability and prior expectations that are based on long-term learned biases (the width–height illusion; Ganel & Goodale, 2003). This avoided possible differences between those with and without autism in learning rates of the prior buildup throughout the experiment. Participants performed a 2AFC task in which they indicated which of two simultaneously presented rectangles was wider. The perceptual bias (prior) was manipulated by comparing trials in which the standard and the comparison rectangles were equal in height to trials in which the comparison rectangle was taller than the standard rectangle. The effect of the prior is typically reflected by a biased performance in perceiving taller rectangles as thinner (Ganel & Goodale, 2003). To manipulate sensory input, we blurred the vertical contours of the rectangles by applying a horizontal Gaussian blur filter at varying degrees of strength (by varying the standard deviations of the filter). This resulted in contours with either sharp, mildly blurred, or highly blurred vertical edges (Figure 1(a)). We used the method of constant stimuli, with individual psychometric functions computed for each condition. The standard rectangles were of constant width and height, and the comparison had two possible heights. For each of the heights, we presented a set of comparisons wider and thinner than the standard. The bias was measured by considering the differences in points of subjective equality (PSEs) for the two comparison heights, reflecting the discrepancy between the perceived width of the taller rectangle and that of the shorter rectangle (Figure 1(b)).

Figure 1.

(a) A sequence of events in a trial. The paradigm follows a constant stimulus procedure in which a standard rectangle is presented simultaneously with a comparison rectangle of varying width, of either same height (a1) or different height (a2). On some of the trials, the vertical edges of the rectangles were blurred by a Gaussian filter at varying degrees (a1). (b) Illustrations depicting relations between Bayesian parameters and psychometric function. (b1) Strength of the illusion affects the position of distribution of judgments and results in shifted psychometric functions (shifts in PSEs). (b2) Reduced sensitivity results in changes in slope (JNDs) and position (PSEs) of the psychometric function. Increasing effects of the illusion are expected for noisier measurements (after Stocker & Simoncelli, 2006).

As predicted by the Bayesian approach, we expected an increase in the bias for TD participants with increasing noise (blur of the vertical contours). A non-negligible bias was expected for TD even in the no-noise condition (sharp contours), demonstrating the involuntary nature of the illusory effect on typical perception. This was further validated by a speeded classification task, measuring the interference between the height and width dimensions. We had two main hypotheses. First, the reduced utilization of priors hypothesis predicts an overall reduced width–height illusion in ASD across the different degrees of blur. This would be manifested in an overall smaller difference in bias, that is, a smaller difference between PSEs of same-height and different-height trials. The second hypothesis posits that the weighting of prior knowledge in relation to the sensory input might differ in ASD, and if so, regardless of the general magnitude of the bias, it would not increase with increased uncertainty in the sensory evidence. Although not mutually exclusive, the two hypotheses point to different underlying mechanisms of modulation in perceptual inference in ASD.

Method

Participants

In total, 70 adults participated in the experiment: 49 TD adults (18 males; mean age = 26.19, range = 20–44) and 21 adults diagnosed with high-functioning ASD (19 males; mean age = 27.48, range = 19–38). Participants with ASD were recruited from community housing, and university students were recruited as TD subjects. Specific data on socioeconomic status and ethnicity were not recorded. All participants (or participants’ guardians) gave written informed consent. The study was approved by the Ethical Committee of the Faculty of Education at the University (Approval Number: 20/046). All participants reported normal or corrected-to-normal vision and had normal or above normal IQ. IQ was assessed using the Test of Nonverbal Intelligence–Fourth Edition (TONI-4, an age-standardized test with a mean score of 100 and a standard deviation of 15; Brown et al., 2010; Ritter et al., 2011). No significant difference in IQ was found between the groups, t(68) = 0.25, p > 0.80 (ASD: mean = 103.00, SD = 10.25, range = 83–124; TD: mean = 103.69, SD = 11.07, range = 86–123). The Autism Spectrum Quotient (AQ; Baron-Cohen et al., 2001) was used to measure autistic traits. AQ scores in the ASD group¹ (mean = 24.20, SD = 5.74) were significantly higher than those in the TD group (mean = 14.63, SD = 7.29), t(67) = 5.24, p < 0.001.

Stimuli and procedure

Participants were tested individually in a dimly lit, sound attenuated room, seated at a distance of 57 cm from a computer screen. Stimuli were generated and presented using MATLAB (Version R2017b; The MathWorks, Inc.) and Psychtoolbox (Brainard, 1997). Two tasks were administered to each participant: the 2AFC width discrimination task and a speeded classification task.

Two-alternative width discrimination task

In each trial, two filled dark gray rectangles, a standard and a comparison, were presented simultaneously against a bright gray background (Figure 1(a)). From a viewing distance of 57 cm, the standard rectangle subtended 2.495° in width and 4.1° in height, and the comparison rectangles varied symmetrically around the standard width, ranging between 1.497° and 3.493° (in increments of 0.11°). Each of the resulting 19 standard-comparison pairs was repeated 24 times. In half of the trials, the height of the comparison rectangle was identical to the standard (4.1°), and in the other half, the comparison was taller (10.9°). To manipulate the amount of noise in the sensory input, a horizontal Gaussian blur filter was applied to the rectangles, creating three degrees of blur of the vertical edges of the rectangles: no blur (no filter), mild blur (Gaussian filter with standard deviation of 0.277°), and strong blur (filter with standard deviation of 0.416°). A total of 1368 trials were administrated for the three degrees of blur. The 456 trials in each blur strength were randomly divided into 6 blocks of 76 trials each, and the resulting 18 blocks were presented in random order, separated by short breaks. In each trial, a 300-ms fixation preceded the 600-ms presentation of the stimulus (rectangle pair). Rectangles were positioned with inner edges at a distance of 2.77° from and on either side of the fixation point. The position of the standard was randomized for the two sides in each trial; it was in the same position for no more than three consecutive trials. Participants were instructed to indicate which of the rectangles was wider. The fixation of the following trial appeared 500 ms after the participant’s response (see Figure 1(a)). Completion of the experiment, including the AQ and IQ tests, required approximately 1.25 h.

Data analysis

Individual sigmoid psychometric functions were fitted for each combination of height and blur using a generalized linear regression model with a logit link. This resulted in six functions per subject. The proportions of “comparison wider” responses were plotted against the width of the comparison rectangle (see Figure 2 for the fitted functions across subjects in each of the groups and supplementary#1 for individual fitted functions). Two measures were extracted from the individual psychometric functions: (1) the PSE, calculated as the rectangle width corresponding to the probability of 50% “comparison wider” responses, and (2) the discrimination threshold or Just Noticeable Difference (JND), calculated as half the difference between the 25% and 75% of “comparison wider” responses (in width). The difference in PSEs between “same-height” and “different-height” trials (the bias) indicated the magnitude of the illusion, and the difference in JNDs (discrimination thresholds) indicated the difference in perceptual sensitivity.

Figure 2.

Psychometric functions. The proportion of trials in which the participants reported the comparison as wider is plotted as a function of the width of the comparisons. Fitting is shown here across subjects in each group. The effect of the illusion is indicated by the shift in the PSEs of the psychometric function of trials in which the comparison’s height is identical to that of the standard (Comp.Equal) and the psychometric function of trials where comparison is taller (Comp.Taller). As predicted, larger shifts in PSEs, indicating a larger bias, are shown in the blurred compared to the no blur condition in TD. In contrast, ASD subjects show a similar shift in PSEs of the functions across the different degrees of blur (denoted NoBlur and StrongBlur). Goodness of fit measured in mean absolute deviations was similar for both groups; TD: 0.0434; ASD: 0.0438.

Speeded classification task

Participants also completed the Garner speeded classification task (Ganel & Goodale, 2003; Garner, 1978), in which they were asked to judge whether the width of a presented rectangle was “wide” or “thin.” Rectangle height was defined as the irrelevant dimension, and participants were asked to ignore its variability. In the baseline condition, the height of the rectangles remained constant, and in the filtering (biasing) condition, the height of the rectangles varied randomly. The rectangle was either 2.218° or 2.495° in width and 4.1° or 8.2° in height. Faster response times and higher accuracy rates in the baseline than in the filtering condition indicated interference from the irrelevant dimension.

Each trial initiated with a fixation point presented for 300 ms followed by a 600-ms presentation of the rectangles. A training session of 18 trials given before the actual testing allowed participants to become acquainted with the two rectangle widths. The experiment consisted of 160 trials divided into four blocks of 40 trials each: two baseline blocks (one for each rectangle height), followed by two filtering blocks or vice versa (randomly chosen). Within each block, the two baseline or the four filtering rectangles were presented randomly, with the restriction of no more than three consecutive trials of the same rectangle.

Data analysis

Mean reaction times (RTs) for correct responses, proportion of correct responses, and an “Inverse Efficiency Score” (IES; Townsend & Ashby, 1978), computed as mean RT divided by the proportion of correct responses, were collected for each participant in each of the conditions. These served as the dependent variables. RTs longer than 5 s were omitted from further analyses (<1% from all correct trials; in the TD group, mean omitted trials were 0.44% (SD = 1%), and in the ASD group, mean omitted trials were 2.22% (SD = 3.01%)). All RTs were longer than 250 ms. A normalized interference score indicating the effect of variability of the irrelevant dimension on performance was also computed using equation (1)

Interference = \frac{Filtering (Accuracy) - Baseline (Accuracy)}{Baseline (Accuracy)}

(1)

Community involvement

There is no community involved in this study.

Results

Figure 2 depicts the fitted psychometric functions across participants in each of the groups. The mean PSEs and JNDs derived from the individual functions are depicted in Figures 3 and 4 as a function of rectangle height and blur strength. The effect of the illusion can be seen for both subject groups (TD and ASD), as indicated by the increase in the PSEs (bias) between the psychometric function of trials where comparison was taller than the standard and the psychometric function of trials with standard and comparison of identical height. A larger bias appeared in the blurred than in the no blur condition in TD participants. In contrast, ASD participants showed a similar bias across the degrees of blur. We first report the analysis of variance (ANOVA) results on the PSEs (perceptual bias) and JNDs (perceptual sensitivity) of the two subject groups, followed by cluster analysis identifying sub-clusters within each of the groups.

Figure 3.

Black lines represent the mean PSEs for trials in which the comparison equaled the standard in height, and gray lines represent trials in which the comparison was taller than the standard. (a) TD participants demonstrated the expected increase in magnitude of the illusion with increase in strength of blur. (b) Among participants with ASD, however, the magnitude of the illusion remained fairly constant and did not scale with the degree of blur.

Figure 4.

Mean JNDs as a function of blur and height for TD and ASD. Results show no significant difference in JNDs between the groups: (a) TD participants demonstrated the expected linear trend of JND increase as a function of noise degree (i.e. decrease in perceptual sensitivity). (b) Among participants with ASD, however, the same trend was found albeit weaker when the comparison equaled the standard in height, and non-linear when the comparison was taller than the standard.

Analysis of PSEs: perceptual biases as a function of stimulus noise

We performed a mixed-design ANOVA on the perceived width of the standard rectangle (mean PSEs), with comparison height (same/different height than the standard) and strength of blur (no blur, mild blur, strong blur) as within-subject factors, and group (TD vs ASD) as the between-subject factor. As hypothesized, we found a significant interaction between comparison height and blur strength, F(2, 68) = 18.15, p < 0.001, $η_{p}^{2}$ = 0.21, indicating an increase in the shift in PSEs between comparison heights as a function of blur strength. This indicates an increase in the bias with increasing blur of the rectangles. Importantly, the significant effect of comparison height, F(1, 68) = 108.15, p < 0.001, $η_{p}^{2}$ = 0.61, was not qualified by group, F(1, 68) = 0.06, p > 0.80, indicating no significant difference in the overall magnitude of the width–height bias between ASD and TD participants. The strength of blur had an effect on the perceived width, F(2, 68) = 16.98, p < 0.001, $η_{p}^{2}$ = 0.20, but the effect varied for the groups, F(2, 68) = 5.33, p < 0.006, $η_{p}^{2}$ = 0.07, with ASD participants demonstrating an overall reduced effect of blur strength on the perceived width. No main effect of group was found, F < 1.

Most importantly, there was a significant interaction between comparison height, strength of blur, and group, F(2, 68) = 3.55, p < 0.031, $η_{p}^{2}$ = 0.05, indicating a difference between TD and ASD in the scaling of the perceptual bias with increased sensory noise. Although the interaction between blur strength and comparison height was observed in both groups, F(2, 48) = 27.92, p < 0.001, $η_{p}^{2}$ = 0.37, and F(2, 20) = 3.56, p < 0.038, $η_{p}^{2}$ = 0.15, for TD and ASD, respectively,² trend analysis performed separately for each height condition demonstrated a different scaling of the perceptual bias for each group. As shown in Figure 3(a), TD participants show a significant linear increase in PSEs with blur strength when the comparison was taller than the standard, F(1, 48) = 65.62, p < 0.001, $η_{p}^{2}$ = 0.58, but not when the comparison and the standard were equal in height, F(1, 48) = 0.23, p > 0.63. All pairwise comparisons of the linear component were significant (ps < 0.004) when the comparison was taller than the standard. For ASD participants, the linear increase in PSEs with blur strength was only marginal when the comparison was taller, F(1, 20) = 4.19, p = 0.054, $η_{p}^{2}$ = 0.17,³ and non-significant when the comparison and the standard were of equal height, F(1, 20) = 0.54, p > 0.47 (see Figure 3(b)). In contrast to the TD group, all pairwise comparisons of the different blur strengths were not significant (p > 0.28, p > 0.99) for the ASD group. Thus, a substantial increase in the magnitude of the bias was shown, as expected, for TD, whereas for ASD, the magnitude of the bias remained fairly constant and did not scale to the same degree as it did for the TD group across the different degrees of blur.

Analysis of JNDs: perceptual sensitivity as a function of stimulus noise and perceptual bias

We performed a mixed-design ANOVA on the discrimination thresholds (mean JNDs) with comparison height (same/different height than the standard) and strength of blur (no blur, mild blur, strong blur) as within-subject factors and group (TD vs ASD) as a between-subject factor. The analysis demonstrated the expected increase in JNDs (decreased perceptual sensitivity) with increasing degree of blur strength, F(2, 68) = 47.13, p < 0.001, $η_{p}^{2}$ = 0.41. In addition, a significant main effect of comparison height, F(1, 68) = 76.15, p < 0.001, $η_{p}^{2}$ = 0.53, indicated that the presence of the illusion reduced the participants’ sensitivity to the rectangle width. A significant interaction between comparison height and blur strength, F(2, 68) = 3.42, p < 0.035, $η_{p}^{2}$ = 0.05, and a significant interaction between comparison height, blur strength, and group, F(2, 68) = 4.12, p < 0.018, $η_{p}^{2}$ = 0.06, indicated that the combined effect of perceptual bias and noise in the sensory input scaled differently for each group. No overall differences were found between the groups, F(2, 68) = 0.07, p > 0.79.

Separate analyses for each group revealed a significant interaction between comparison height and blur strength for TD participants, F(2, 48) = 4.95, p < 0.001, $η_{p}^{2}$ = 0.09. A trend analysis demonstrated stronger linear trends of JND increase as a function of noise degree (i.e. decrease in perceptual sensitivity) when rectangles were equal in height, F(1, 48) = 116.55, p < 0.001, $η_{p}^{2}$ = 0.71; there was a significant, albeit somewhat weaker, linear trend when the comparison rectangle was taller, F(1, 48) = 24.12, p < 0.001, $η_{p}^{2}$ = 0.33. Thus, as shown in Figure 4(a), the effect of the illusion on perceptual sensitivity decreased as the stimulus became noisier. For ASD participants (Figure 4(b)), the interaction between comparison height and blur strength was marginally significant, F(1, 20) = 3.17, p = 0.053, $η_{p}^{2}$ = 0.14. A trend analysis revealed the expected linear trend of JND increase as a function of noise strength (i.e. decrease in perceptual sensitivity) when the rectangles were equal in height, F(1, 20) = 16.96, p < 0.001, $η_{p}^{2}$ = 0.46; however, when the comparison rectangle was taller, the dominant trend was quadratic, F(1, 20) = 11.33, p < 0.003, $η_{p}^{2}$ = 0.36.

Clusters of performance among TD participants

To examine individual differences in the bias magnitude (i.e. susceptibility to the width–height illusion) and its scaling with noise in the stimulus, we applied a two-step cluster procedure to the TD data, on three continuous variables indicating the individual bias (differences in PSE values between the two comparison heights) for each of the three degrees of blur. Larger differences indicated a larger effect of the illusion. We used the Schwarz Bayesian information criterion (BIC; see Schwarz, 1978) to determine subgroups based on the initial bias and on the scaling of the perceptual bias as a function of the degree of noise in the stimulus. The algorithm defined two clusters as best fitting the data; the lowest Schwarz BIC was 108.19, and the highest ratio of distance measures was 3.91. The first cluster contained 63.3% (31/49) of the TD sample, and the second cluster contained the remaining 36.7% (18/49).

We performed a mixed-design ANOVA on the PSEs of the TD samples with cluster as a between-subject factor, and comparison height (same/different height) and the three degrees of blur as the within-subject factors. To avoid repetition, we only report the results for the cluster factor. The analysis revealed a significant effect of cluster, F(1, 47) = 33.99, p < 0.001, $η_{p}^{2}$ = 0.42, validating the identified clusters. Cluster interacted significantly with comparison height, F(1, 47) = 102.52, p < 0.001, $η_{p}^{2}$ = 0.69, indicating that the two identified clusters differed in their overall susceptibility to the illusion, but critically, there was no interaction between cluster, height, and blur, F(1, 47) = 1.28, p > 0.28, indicating that despite large initial differences in susceptibility to the illusion, the pattern of increasing perceptual bias as a function of noise was similar for the two clusters (see Figure 5(a) and (c)). Moreover, no difference between clusters was found in TONI and AQ (Fs < 1).

Figure 5.

Black lines represent the mean PSEs for trials in which the comparison equaled the standard in height, and gray lines represent trials in which the comparison was taller than the standard. In both groups, Cluster 1 (a, b) consisted of approximately two-thirds of the participants and showed a smaller initial strength of illusion relative to participants in Cluster 2 (c, d). Nevertheless, both clusters of TD participants (a, c) demonstrated the expected increase in magnitude of the illusion with increase in degree of blur. In contrast, both clusters of participants with ASD showed a constant magnitude of the illusion with only weak scaling of the illusion with increasing degrees of blur (b, d).

Clusters of performance among ASD participants

Following the same protocol as above for TD participants, we applied the clustering procedure to the ASD group. The algorithm defined two clusters that best fitted the data; the lowest Schwarz BIC was 59.78, and the highest ratio of distance measures was 2.45. The first cluster contained 61.9% (13/21) of the ASD sample, and the second contained the remaining 38.1% (8/21).

We performed a mixed-design ANOVA on the PSEs of the ASD sample, with cluster as a between-subject factor and comparison height (same/different height) and blur strength as within-subject factors. The analysis revealed a significant effect of cluster, F(1, 19) = 31.03, p < 0.001, $η_{p}^{2}$ = 0.62, that, as observed for the TD individuals, interacted significantly with the comparison height, F(1, 19) = 42.26, p < 0.001, $η_{p}^{2}$ = 0.69, demonstrating a general difference between the two clusters in susceptibility to the illusion. However, unlike the TD individuals, the pattern of increased perceptual bias as a function of noise was not seen in either of the two identified clusters (see Figure 5(b) and (d)). The interaction between cluster, height, and blur was marginally significant, F(2, 19) = 2.60, p = 0.088, $η_{p}^{2}$ = 0.12, and separate analysis for each cluster showed no interaction between comparison height and blur strength for the ASD participants comprising Cluster 1, F(1, 12) = 0.43, p > 0.65, indicating that the bias remained fairly constant in magnitude across the different degrees of noise in the measurement. No linear trend was found for the equal height trials, F(1, 12) = 0.008, p > 0.93, or for the different-height trials, F(1, 12) = 1.34, p > 0.27. In contrast, for ASD participants in Cluster 2, the interaction between comparison height and blur strength was significant, F(1, 7) = 4.08, p < 0.040, $η_{p}^{2}$ = 0.37; however, no linear trend was observed for the equal height trials, F(1, 7) = 0.84, p > 0.39, or for the different-height trials, F(1, 7) = 3.16, p > 0.11, indicating weak scaling of the perceptual bias as a function of noise in the measurement. Similar to the TD individuals, the clusters of ASD showed no difference in TONI and AQ (Fs < 1).

Comparing the cluster performances of TD and ASD participants

We further examined whether the difference found in the relative weighting between the groups was preserved regardless of the overall illusion magnitude displayed by individuals in each cluster.

We compared the two clusters within each of the groups using a mixed-design ANOVA on PSEs, with group and cluster as between-subject factors and comparison height (same/different height than the standard) and blur strength (no blur, mild blur, strong blur) as within-subject factors. To avoid repetition, only results involving interactions between group and cluster are reported here. The analysis revealed a significant interaction between cluster, group, comparison height, and blur strength, F(2, 66) = 3.13, p < 0.047, $η_{p}^{2}$ = 0.05. Within each of the TD and ASD groups, the two identified clusters differed in the initial susceptibility to the illusion (for the no blur condition), yet both clusters in the TD group showed clear scaling of the bias, whereas ASD participants showed no such scaling, for subjects in Cluster 1 (exhibiting low susceptibility to the illusion), and weak scaling for those in Cluster 2 (exhibiting high susceptibility to the illusion). This pattern of results suggests individuals with ASD differ in the scaling of perceptual bias, regardless of their basic susceptibility to the illusion.

Interference between width and height—speeded classification task

Results from the Garner speeded classification task, in which participants were asked to judge whether the width of a presented rectangle was “wide” or “thin,” are shown in Figure 6. Mean RTs, IES, accuracy rates, and normalized interference scores are plotted as a function of condition and group.

Figure 6.

The difference between baseline blocks (black) and filtering blocks (gray) demonstrates the large effect of the irrelevant dimension on classification in all measures and among both groups; TD showed overall faster responses (RT, left panel of a), better efficiency (IES, right panel of a), and accuracy (b). Most importantly, in terms of efficiency and accuracy, this difference was significantly greater in the ASD group, indicating greater interference of the irrelevant dimension, even after normalizing to the baseline RTs (c).

Accuracy and RT analyses

We performed a mixed-design ANOVA with condition (baseline vs filtering) as a within-subject factor and group (TD vs ASD) as a between-subject factor on the mean correct RTs and accuracy rates (see Figure 6). The analysis revealed a significant effect of condition on both mean RTs and accuracy rates, F(1, 68) = 49.44, p < 0.001, $η_{p}^{2}$ = 0.42 and F(1, 68) = 65.19, p < 0.001, $η_{p}^{2}$ = 0.49, respectively, demonstrating the cost of the irrelevant dimension on classification speed and accuracy. Importantly, for accuracy rates, this effect of condition interacted with group, F(1, 68) = 4.51, p < 0.037, $η_{p}^{2}$ = 0.06, indicating greater interference in the ASD group. No group differences were found for the baseline condition, t(68) = 0.38, p > 0.70, indicating group differences most likely originate in the filtering condition. This was confirmed in a t-test analysis of the accuracy rates computed as interference scores normalized to the baseline performance (see equation (1) in the “Method” section), t(68) = 2.29, p < 0.025. This interaction was not found for RTs, F(1, 68) = 1.19, p > 0.28, but ASD participants were overall slower in responding, F(1, 68) = 17.60, p < 0.001, $η_{p}^{2}$ = 0.21.

IES analysis

The analysis of IES scores revealed a similar pattern. A significant effect of condition was found, F(1, 68) = 90.69, p < 0.001, $η_{p}^{2}$ = 0.57, demonstrating the cost of the irrelevant dimension (rectangle height) on classification performance, and a significant effect of group, F(1, 68) = 19.12, p < 0.001, $η_{p}^{2}$ = 0.22, indicating a lower IES and an overall lower performance among ASD participants. Importantly, a significant interaction between condition and group, F(1, 968) = 8.02, p < 0.006, $η_{p}^{2}$ = 0.11, demonstrated, once again, a larger interference of the irrelevant dimension in the ASD group.

Discussion

Many previously unexplained systematic biases in perception have been proposed as reflecting Bayesian reliance on prior knowledge (e.g. Alais & Burr, 2004; Stocker & Simoncelli, 2006). There is also evidence that the nervous system reduces the overall uncertainty of estimates by combining sensory measurements with prior knowledge and biases. The ideal Bayesian observer changes the relative reliance on the sensory measurement and on the prior knowledge, as the relative certainty of each changes. Here, we examined the effects of the reliability of the sensory input on width judgments by systematically introducing noise in the stimulus. Consequently, the relative reliability of the prior was expected to increase, leading to more strongly biased estimates.

This pattern was strongly demonstrated in our data in TD with significantly increased magnitude of the width–height illusion for noisier stimuli. Interestingly, individuals with autism demonstrated similar susceptibility to the illusion and comparable thresholds to those of the TD individuals. However, the relative weighting of the prior and the sensory input was atypical; they displayed non-adaptive, consistent biases across the different degrees of sensory noise, in clear contrast to the monotonically increasing biases displayed by the TD. These findings provide clear evidence of the utilization of prior knowledge in the perceptual judgments of individuals with autism. Importantly, however, these findings also suggest that the main modulation in ASD results from inflexibility in adjusting the prior in accord with the increased uncertainty of the sensory input. Post hoc cluster analyses showed that this difference between the groups in adjusting the prior was preserved regardless of individual differences in the illusion magnitude displayed under the low level of stimulus uncertainty (no blur condition). While the two identified clusters within each group differed in their initial susceptibility to the illusion, both clusters of the TD group showed clear scaling of the bias, whereas both clusters of the ASD participants showed non-adaptive scaling. ASD individuals in Cluster 1 (exhibiting low susceptibility to the illusion) as well as those in Cluster 2 (exhibiting high susceptibility to the illusion) barely displayed any scaling of bias, suggesting non-adaptive relative weighting of the prior and sensory reliability, regardless of the individual’s basic susceptibility to the illusion. The study thus tapped central differences in the inferred perception of TD individuals and ASDs that were consistent across the identified sub-clusters within each group.

The strong claim of attenuated Bayesian priors in autism cannot account for the results. According to this view, more veridical perception is predicted for ASD, with representations remaining close to the sensory input as a consequence of the weak priors; however, the bias measured in our study (under the no blur condition) did not differ in magnitude for the different groups, implying priors were not necessarily weaker in ASD participants. Instead, the weighting of the sensory input against the prior knowledge and experience in determining the final percept was modulated in the ASD group. Specifically, prior expectations, based in our case on long-term learned biases (the width–height illusion), were not adjusted to the measurement reliability on the width discrimination task. This reduced flexibility in autism in adjusting the weighting of the priors has been demonstrated in several recent studies, indicated by an overall slower learning of prior information (Soulières et al., 2011), slower updating of the prior by recent history of stimuli (Lieder et al., 2019), and a failure to flexibly adjust the prior precision to the context (Sapey-Triomphe et al., 2020). However, in all of these studies, differences in adjusting the priors and updating the relative reliance on the measurements for those with and without autism may be explained by potential differences in the learning rates of building the prior throughout the experiment. We used long-term learned biases as priors while manipulating uncertainty in the stimulus. The results are therefore unlikely to reflect slower rates of building up a precise representation of the prior in autism and can be taken as direct evidence of reduced flexibility in adjusting the role of prior knowledge versus that of sensory evidence.

Individuals with ASD have also been shown to exhibit inflexibility in the way they calibrate their sensitivity to changes in the incoming input. In both visual discrimination and in somatosensory discrimination tasks, ASDs do not seem to scale perceptual sensitivity (i.e. JNDs) with the magnitude of the stimulation, as defined by Weber’s law (Hadad & Schwartz, 2019). This violation of Weber’s law in autism demonstrates disruptions in auto-calibrating the perceptual system to the environment which normally makes maximal use of the limited processing range of the system, disregarding minor and insignificant changes in the input while amplifying sensitivity to significant information. Changes in the stimuli are registered in autism with fairly constant noise in the encoded incoming input, in an absolute and inflexible manner, with no immediate calibration based on the stimulation itself (Rakitin et al., 1998).

Discrimination thresholds in both TD and ASD groups increased with increasing noise in the stimuli. In fact, both higher noise levels introduced to the contours (in the width discrimination task) and increased variability of the irrelevant dimension (in the Garner task) impaired perceptual judgments in both groups. If anything, the irrelevant dimension (rectangle’s height) impaired performance more in the ASD group, presumably because of participants’ weaker selective attentional skills (e.g. Burack, 1994). Thus, the modulations found in the adjustments of the prior to noise levels in the ASD group cannot be explained by overall differences in perceptual sensitivity to noise.

Our results also demonstrate clear susceptibility to perceptual illusions in those diagnosed with autism. Previous studies have demonstrated reduced effects (e.g. Mitchell et al., 2010) and, in some cases, the complete resistance of individuals with autism to perceptual illusions (Happé, 1996; Simmons et al., 2009). However, as recently shown for illusions incorporating the integration of multisensory input (Hadad & Schwartz, 2019) and perceptual organization (Avraam et al., 2019), when tested psychophysically with full psychometric functions, those with and without autism show no difference in the overall magnitude of the perceptual bias. Moreover, these illusive effects on the interpretation of incoming input seem robust and mandatory in autism and, as in neurotypicals, affect perceptual judgments even under relatively low uncertainty in the incoming stimulus. These findings highlight the importance of the psychophysical measures employed in testing perceptual representations in general and in those with autism in particular.

Susceptibility to perceptual illusions has often been assessed with tasks in which the participants are required to introspect and explicitly report on their own perception. Differences between groups could thus reflect other trivial differences than those related to the specific perceptual process under study. Susceptibility to illusions can be more adequately inferred based on implicit measurements of the perceived magnitude of stimuli, extracted from the full psychometric functions of perceptual discrimination tasks. As in the current data, comparable effects of perceptual illusions on those with and without autism are observed in such cases. These results point to the importance of task factors in the study of perception in ASD.

What is then the possible underlying model of the modulated perception we observed in ASD? We find that neither the reduced prior nor reduced sensitivity to noise in the sensory input are sufficiently supported. We propose a possible model that relies on classic Bayesian modeling in human perception and introduce a second-order factor. In the proposed model, second-order beliefs on the precision itself are represented to account for the differences in adjustments of the relative weighting of perceptual bias and sensory reliability. Similar ideas have been proposed by the predictive coding explanation (Friston et al., 2013), suggesting the abnormality in autism is not a failure of prediction per se, but a failure to instantiate top-down predictions during perceptual synthesis. By this account, there is a failure of beliefs (estimated precision) about beliefs (predictions), that is, a failure of a second-order metacognitive computation. Our suggested model incorporating second-order behavior beyond the Bayesian belief is described in Appendix 1. However, future studies are needed to determine the role of this factor in relation to the traditional Bayesian modeling parameters in accounting for autistic perception.

To conclude, we examined the effects of long-term priors on the interpretation of incoming input in TD and ASD, while systematically manipulating uncertainty in the sensory input. We demonstrated clear differences between the groups in the relative weighting of the prior and that of the sensory input in determining the final percept. High-functioning individuals with autism exhibited similar susceptibility to the illusion and comparable thresholds to those of the TD individuals; however, they displayed a non-adaptive, consistent bias across the different degrees of sensory noise. These findings provide clear evidence for the utilization of prior knowledge in the perceptual judgments of those with autism. However, they suggest an inflexible and non-adaptive mechanism of adjustment that may account for the atypical perceptual inference in ASD.

Supplemental Material

sj-docx-1-aut-10.1177_13623613221074416 – Supplemental material for Individuals with autism show non-adaptive relative weighting of perceptual prior and sensory reliability

Supplemental material, sj-docx-1-aut-10.1177_13623613221074416 for Individuals with autism show non-adaptive relative weighting of perceptual prior and sensory reliability by Nahal Binur, Hagit Hel-Or and Bat-Sheva Hadad in Autism

Footnotes

Appendix 1

Classic Bayesian theory states that the posterior distribution is a product of the prior distribution and the likelihood. Most behavioral models that use Bayesian modeling assume that the distributions are all Gaussian distributions.

Let w_stim denote the input stimulus (e.g. width of the rectangle), and denote by P(w|w_stim) the posterior distribution (the final estimate $\hat{w}$ is taken as the mode or argmax of this distribution). Following Bayes’ rule, we have

(2)

\hat{w} = {argmax}_{w} (P (w | w_{stim})) = {argmax}_{w} (P (w) P (w_{stim} | w))

Assuming Gaussian distribution:

(1)

The likelihood distribution : P (w_{stim} | w) ~ N (µ_{s}, σ_{s}),

where µ_s,σ_s are the mean and SD of the normal distribution (e.g. the internal representation of the rectangle width).

(2)

The prior distribution: : P (w) ~ N (µ_{p}, σ_{p}),

where µ_p,σ_p are the mean and SD of the normal distribution (e.g. the illusory width–height effect on the width of the rectangle).

The posterior P(w|w_stim) is then a Gaussian distribution as well, with the following parameters

(3)

μ_{post} = \frac{\frac{1}{σ_{p}} μ_{p} + \frac{1}{σ_{s}} μ_{s}}{\frac{1}{σ_{p}} + \frac{1}{σ_{s}}}

(4)

σ_{post} = \frac{1}{\frac{1}{σ_{p}} + \frac{1}{σ_{s}}}

$\hat{w}$ being the maximum of a Gaussian distribution is given by

(5)

\hat{w} = \frac{\frac{1}{σ_{p}} μ_{p} + \frac{1}{σ_{s}} μ_{s}}{\frac{1}{σ_{p}} + \frac{1}{σ_{s}}}

Thus, $\hat{w}$ is a weighted average of µ_p and µ_s with the weights determined by the uncertainty of the prior and the likelihood (reliability of the width–height illusion and the reliability (noise) of the sensory input).

We suggest that the weights are in fact determined by two factors. The first being the above reliabilities of the prior and the likelihood, and the second is a directive factor introduced by the individual, which assigns a preference and higher weighting for either the prior or the likelihood. We will denote this factor by K and reintroduce the above equation as

(6)

\hat{w} = \frac{K \frac{1}{σ_{p}} μ_{p} + (1 - K) \frac{1}{σ_{s}} μ_{s}}{\frac{K}{σ_{p}} + \frac{(1 - K)}{σ_{s}}}

Note that K could in practice be incorporated into the reliability factor σ; however, this representation allows the distinction between the reliability and the directive factors which most likely have a different behavioral pattern as we describe below. The formulation in equation (6) can assist in modeling the distinction in perceptual behavior between autism spectrum disorder (ASD) and typically developed (TD) individuals.

The suggested model can be viewed as incorporating a second-order behavior beyond the Bayesian belief, thus representing second-order beliefs on the precision itself. By this account, there is a failure of beliefs (estimated precision) about beliefs (predictions), that is, a failure of a second-order metacognitive computation. This should be further examined experimentally. However, due to the confounding nature of the directive factor K and the reliability factors σ_p and σ_s, testing the model cannot rely on static analysis and will require a dynamically changing environment. Only then does the distinction between K and sigma come into play and might possibly be disentangled. Future studies focusing on this possible second-order behavior may shed light on perceptual inference in ASD.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research was funded by the Israel Science Foundation (ISF), Grant #882/19.

ORCID iD

Bat-Sheva Hadad

Supplemental material

Supplemental material for this article is available online.

Notes

References

Alais

Burr

(2004). The ventriloquist effect results from near-optimal bimodal integration. Current Biology, 14(3), 257–262. https://doi.org/10.1016/j.cub.2004.01.029

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (DSM-5®).

Avraam

Binur

Hadad

B. S.

(2019). Typical perceptual organization in autism: Perceptual grouping and spatial distortion. Autism Research, 12(11), 1623–1635. https://doi.org/10.1002/aur.2153

Baron-Cohen

Wheelwright

Skinner

Martin

Clubley

(2001). The Autism-Spectrum Quotient (AQ): Evidence from Asperger syndrome/high-functioning autism, males and females, scientists and mathematicians. Journal of Autism and Developmental Disorders, 31(1), 5–17. https://doi.org/10.1023/A:1005653411471

Brainard

D. H.

(1997). The psychophysics toolbox. Spatial Vision, 10(4), 433–436. https://doi.org/10.1163/156856897X00357

Brock

(2012). Alternative Bayesian accounts of autistic perception: Comment on Pellicano and Burr. Trends in Cognitive Sciences, 16(12), 573–574. https://doi.org/10.1016/j.tics.2012.10.005

Brown

Sherbenou

R. J.

Johnsen

S. K.

(2010). TONI-4: Test of Nonverbal Intelligence–Fourth Edition. Pro-Ed.

Burack

J. A.

(1994). Selective attention deficits in persons with autism: Preliminary evidence of an inefficient attentional lens. Journal of Abnormal Psychology, 103(3), 535–543. https://doi.org/10.1037/0021-843X.103.3.535

Dakin

Frith

(2005). Vagaries of visual perception in autism. Neuron, 48(3), 497–507. https://doi.org/10.1016/j.neuron.2005.10.018

10.

Evers

Van der Hallen

Noens

Wagemans

(2018). Perceptual organization in individuals with autism spectrum disorder. Child Development Perspectives, 12(3), 177–182. https://doi.org/10.1111/cdep.12280

11.

Feigin

Shalom-Sperber

Zachor

D. A.

Zaidel

(2021). Increased influence of prior choices on perceptual decisions in autism. eLife, 10, Article e61595. https://doi.org/10.7554/eLife.61595

12.

Ferri

Elia

Agarwal

Lanuzza

Musumeci

S. A.

Pennisi

(2003). The mismatch negativity and the P3a components of the auditory event-related potentials in autistic low-functioning subjects. Clinical Neurophysiology, 114(9), 1671–1680. https://doi.org/10.1016/S1388-2457(03)00153-6

13.

Friston

K. J.

(2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787

14.

Friston

K. J.

Lawson

Frith

C. D.

(2013). On hyperpriors and hypopriors: Comment on Pellicano and Burr. Trends in Cognitive Sciences, 17(1), 1. https://doi.org/10.1016/j.tics.2012.11.003

15.

Ganel

Goodale

M. A.

(2003). Visual control of action but not perception requires analytical processing of object shape. Nature, 426, 664–667. https://doi.org/10.1038/nature02156

16.

Garner

W. R.

(1978). Selective attention to attributes and to stimuli. Journal of Experimental Psychology: General, 107(3), 287–308. https://doi.org/10.1037/0096-3445.107.3.287

17.

Hadad

B. S.

Goldstein

E. K.

Russo

N. N.

(2017). Atypical perception in autism: A failure of perceptual specialization? Autism Research, 10(9), 1510–1522. https://doi.org/10.1002/aur.1800

18.

Hadad

B. S.

Schwartz

(2019). Perception in autism does not adhere to Weber’s law. eLife, 8, Article e42223. https://doi.org/10.7554/eLife.42223

19.

Hadad

B. S.

Schwartz

Binur

(2019). Reduced perceptual specialization in autism: Evidence from the other-race face effect. Journal of Experimental Psychology: General, 148(3), 588–594. https://doi.org/10.1037/xge0000550.

20.

Happé

F. G.

(1996). Studying weak central coherence at low levels: Children with autism do not succumb to visual illusions. A research note. Journal of Child Psychology and Psychiatry, 37(7), 873–877. https://doi.org/10.1111/j.1469-7610.1996.tb01483.x

21.

Kujala

Aho

Lepistö

Jansson-Verkasalo

Nieminen-von Wendt

Von Wendt

Näätänen

(2007). Atypical pattern of discriminating sound features in adults with Asperger syndrome as reflected by the mismatch negativity. Biological Psychology, 75(1), 109–114. https://doi.org/10.1016/j.biopsycho.2006.12.007

22.

Lawson

R. P.

Mathys

Rees

(2017). Adults with autism overestimate the volatility of the sensory environment. Nature Neuroscience, 20(9), 1293–1299. https://doi.org/10.1038/nn.4615

23.

Lawson

R. P.

Rees

Friston

K. J.

(2014). An aberrant precision account of autism. Frontiers in Human Neuroscience, 8, Article 302. https://doi.org/10.3389/fnhum.2014.00302

24.

Lieder

Adam

Frenkel

Jaffe-Dax

Sahani

Ahissar

(2019). Perceptual bias reveals slow-updating in autism and fast-forgetting in dyslexia. Nature Neuroscience, 22(2), 256–264. https://doi.org/10.1038/s41593-018-0308-9

25.

Manning

Kilner

Neil

Karaminis

Pellicano

(2017). Children on the autism spectrum update their behaviour in response to a volatile environment. Developmental Science, 20(5), Article e12435. https://doi.org/10.1111/desc.12435

26.

Mitchell

Mottron

Soulières

Ropar

(2010). Susceptibility to the Shepard illusion in participants with autism: Reduced top-down influences within perception? Autism Research, 3(3), 113–119. https://doi.org/10.1002/aur.130

27.

Pellicano

Burr

(2012). When the world becomes ‘too real’: A Bayesian explanation of autistic perception. Trends in Cognitive Sciences, 16(10), 504–510. https://doi.org/10.1016/j.tics.2012.08.009

28.

Petzschner

F. H.

Glasauer

Stephan

K. E.

(2015). A Bayesian perspective on magnitude estimation. Trends in Cognitive Sciences, 19(5), 285–293. https://doi.org/10.1016/j.tics.2015.03.002

29.

Rakitin

B. C.

Gibbon

Penney

T. B.

Malapani

Hinton

S. C.

Meck

W. H.

(1998). Scalar expectancy theory and peak-interval timing in humans. Journal of Experimental Psychology: Animal Behavior Processes, 24(1), 15–33. https://doi.org/10.1037/0097-7403.24.1.15

30.

Ritter

Kilinc

Navruz

Bae

(2011). Test Review: L. Brown, R. J. Sherbenou, & S. K. Johnsen Test of Nonverbal Intelligence–4 (TONI-4). Austin, TX: PRO-ED, 2010. Journal of Psychoeducational Assessment, 29(5), 484–488. https://doi.org/10.1177/0734282911400400

31.

Sapey-Triomphe

L. A.

Timmermans

Wagemans

(2020). Less flexible perceptual learning of priors in adults with autism. Journal of Vision, 20(11), Article 520. https://doi.org/10.1167/jov.20.11.520

32.

Schwarz

(1978). The Bayesian information criterion. Annals of Statistics, 6, 461–464.

33.

Simmons

D. R.

Robertson

A. E.

McKay

L. S.

Toal

McAleer

Pollick

F. E.

(2009). Vision in autism spectrum disorders. Vision Research, 49(22), 2705–2739. https://doi.org/10.1016/j.visres.2009.08.005

34.

Soulières

Mottron

Giguère

Larochelle

(2011). Category induction in autism: Slower, perhaps different, but certainly possible. Quarterly Journal of Experimental Psychology, 64(2), 311–327. https://doi.org/10.1080/17470218.2010.492994

35.

Stocker

A. A.

Simoncelli

E. P.

(2006). Noise characteristics and prior expectations in human visual speed perception. Nature Neuroscience, 9(4), 578–585. https://doi.org/10.1038/nn1669

36.

Teufel

Subramaniam

Fletcher

P. C.

(2013). The role of priors in Bayesian models of perception. Frontiers in Computational Neuroscience, 7, Article 25. https://doi.org/10.3389/fncom.2013.00025

37.

Thomas

Nardini

Mareschal

(2010). Interactions between “light-from-above” and convexity priors in visual development. Journal of Vision, 10(8), Article 6. https://doi.org/10.1167/10.8.6

38.

Townsend

J. T.

Ashby

F. G.

(1978). Methods of modeling capacity in simple processing systems. In Castellan

Restle

(Eds.), Cognitive theory (Vol. III, pp. 200–239). Lawrence Erlbaum Associates.

39.

Van de Cruys

Evers

Van der Hallen

Van Eylen

Boets

de-Wit

Wagemans

(2014). Precise minds in uncertain worlds: Predictive coding in autism. Psychological Review, 121(4), 649–675. https://doi.org/10.1037/a0037665

40.

Van de Cruys

Lemmens

Sapey-Triomphe

L. A.

Chetverikov

Noens

Wagemans

(2021). Structural and contextual priors affect visual search in children with and without autism. Autism Research, 14, 1484–1495. https://doi.org/10.1002/aur.2511

41.

Von Helmholtz

(1867). Handbuch der physiologischen Optik [Handbook of Physiological Optics]. Leopold Voss.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.59 MB