A popular model for sensory processing, known as predictive coding, proposes that incoming signals are iteratively compared with top-down predictions along a hierarchical processing scheme. At each step, error signals arising from differences between actual input and prediction are forwarded and recurrently minimized by updating internal models to finally be “explained away”. However, the neuronal mechanisms underlying such computations and their limitations in processing speed are largely unknown. Further, it remains unclear at which step of cortical processing prediction errors are explained away, if at all. In the present study, human subjects briefly viewed the superposition of two orthogonally oriented gratings followed by abrupt removal of one orientation after either 33 or 200 milliseconds. Instead of strictly seeing the remaining orientation, observers report rarely but highly significantly an illusory percept of the arithmetic difference between previous and actual orientations. Previous findings in cats using the identical paradigm suggest that such difference signals are inherited from first steps of visual cortical processing. In light of early modeling accounts of predictive coding, in which visual neurons were interpreted as residual error detectors signaling the difference between actual input and its temporal prediction based on past input, our data may indicate continued access to residual errors. Such strategy permits time-critical perceptual decision making across a spectrum of competing internal signals up to the highest levels of processing. Thus, the occasional appearance of a prediction error-like illusory percept may uncover maintained flexibility at perceptual decision stages when subjects cope with highly dynamic and ambiguous visual stimuli.
Citation: Staadt R, Philipp ST, Cremers JL, Kornmeier J, Jancke D (2020) Perception of the difference between past and present stimulus: A rare orientation illusion may indicate incidental access to prediction error-like signals. PLoS ONE 15(5): e0232349. https://doi.org/10.1371/journal.pone.0232349
Editor: Guido Hesselmann, Psychologische Hochschule Berlin, GERMANY
Received: October 25, 2019; Accepted: April 14, 2020; Published: May 4, 2020
Copyright: © 2020 Staadt et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation - Project ID 122679504 - SFB 874 (project part A2) to DJ, German-Israeli Project Cooperation (DIP, JA 945/3-1, SL 185/1-1) to DJ, Deutsche Forschungsgemeinschaft, SPP 1665 (JA 945/4-1) to DJ, and the Bundesministerium für Bildung und Forschung, BMBF. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Visual input is permanently changing and ambiguous. This puts demands on perceptual decisions to be fast and coherent, performing a tightrope act between stability and flexibility. However, continued visual stimulation and detected regularities therein allow predictions, and consequently, promote transmission of attenuated error signals given that top-down predictions are iteratively reconciled with bottom-up input in a hierarchical recurrent network [1–4]. Hence, the observation that repeated visual stimuli lead to reduced cortical activation during context-dependent cognitive tasks involving expectations and attention (for review see ) has been interpreted as an outcome of "predictive coding" at various stages of processing [4,6–13]. While it remains an open question whether reduction of visual responses to repeated or prolonged input may strictly emerge from computation of repetition probabilities  rather than arising from specific local adaptation dynamics [14–16], it has frequently been shown that abrupt alterations in temporal sequences of sensory input produce change-related signals that are generated “automatically” at early steps of cortical processing and that are typically labeled as prediction errors (for review see ). Basic forms of predictive coding have been proposed to exist already at earliest steps of visual processing in the retina  and LGN [19,20] where, in the framework of information theory , center-surround antagonism within receptive fields allow reduction in redundancies by removing predictable components of visual input. Mechanistically, exploiting the spatial (or temporal) correlations within natural images, the surround generates a statistical prediction of the signal at the center that is subtracted from the actual center signal to protect against intrinsic noise . This type of predictive coding was refined for visual cortical neurons using a hierarchical model in which residual errors (signaling the difference between an input signal and its spatiotemporal prediction) propagate via feedforward connections and are compared at each level to correct its current estimate and then update feedback predictions, thereby iteratively decorrelating (i.e., whitening) actual input based on statistical regularities of natural scenes . However, apart from detailed modeling mechanisms, the temporal constrains of predictive coding and to which extent residual error signals are passed through the visual hierarchy are less understood. Here we used temporal sequences of two stimuli comprising gratings with different orientations. Followed in rapid succession these stimuli have been shown to generate robust “prediction error-like” signals in primary visual cortex of anesthetized cats , thus, producing responses without influence of higher level expectation . In the present study we investigated, whether we can find indications of residual prediction errors at such fast time scales in human perception.
Materials and methods
In total, 15 volunteers (age range: 18–26 years) participated in this study. Participants reported normal or corrected to normal vision (as further verified by the Freiburg visual acuity test (FrACT, ). They were naïve to the hypotheses and goals of the experiment (instructions and the task were introduced in a written form) and paid for their participation. All participants provided written informed consent prior to the start of the study. The study was approved by the local Ethical Committees of the University of Freiburg and carried out in accordance with the Declaration of Helsinki.
Experiments were set up and run using custom-written Python scripts. Stimuli were displayed on a TFT screen (NEC, Model PA241W-BK) and viewed at 60 cm distance. Stimuli were circular windowed sine wave gratings (size = 4.8°, spatial frequency = 2.34 cycles/°) presented on dark background (0.4 cd/m2) with peak luminance of 376 cd/m2. The test stimulus was a windowed sine wave grating (size = 4.8°) presented in pseudo-randomized order at various orientations (-90° to 90° in steps of 15°) with spatial frequency of 1.28 cycles/° and peak luminance of 0.8 cd/m2.
The experiment consisted of 3 control conditions and 2 test conditions. After a brief acoustic signal, a fixation dot (Gaussian, FWHM: 0.17°) appeared in the center of the screen for 500 milliseconds. Control conditions: In one (“pure”) control condition, to measure baseline performance, a probe stimulus (Fig 1iii) was presented directly after the fixation dot for 500 milliseconds. In two further control conditions, gratings of oblique orientations were superimposed and presented with different durations, for either 33 ms or 200 milliseconds (Fig 1i). Test conditions: Superimposed gratings (with the same above 2 durations, presented in separate blocks) were followed by a single grating (Fig 1ii) comprising one of the orientations present in the preceding superposition (+45° or -45°, in randomized order). Just as in the pure control condition, in all conditions a probe stimulus appeared (Fig 1iii) followed by a mask (windowed white noise). After 100 ms a response bar was shown (Fig 1iv, single cosine wave, spatial frequency 1.07 cycles/°, in a Gaussian envelope, FWHM 2.07°) at one of 8 random locations next to the mask (distance to fixation point = 6.2°). Participants were requested to rotate the bar using a dial until the bar matched the perceived orientation of the test stimulus. After the response the next trial started.
Test condition: After fixation, (i) superimposed gratings were shown for 33 or 200 ms followed by (ii) a single grating of either -45° or +45° orientation (i.e., matching to one of the orientations shown in the superposition before, size of stimuli = 4.8°). (iii) Thereafter, a probe stimulus was presented for 500 ms (iv.) Participants were asked to rotate a response bar until it matched the perceived orientation of the previous probe stimulus. Control conditions: In “Control Condition 1” only the probe stimulus (iii) was presented after fixation. In two further controls the superposition was presented at both durations, while omitting subsequent presentation of the single grating stimuli (ii).
In each experimental condition 13 different probe orientations were tested. Each such trial was repeated 8 times (i.e., subjects performed 104 perceptual decisions per condition). Participants were instructed to respond as fast and accurate as possible, while keeping fixation of the dot for the duration of each trial.
After fixation (Fig 1, top), human participants (n = 15) viewed a superposition of two orthogonal gratings (Fig 1i) for either 33 ms or 200 ms presentation time. The superposition grating was then followed by a single grating comprising one of the previously superimposed orientations (Fig 1ii, -45° or +45°). Thus, participants viewed a simple switch from a superposition grating to a single grating with two different time intervals. Thereafter, a probe stimulus, containing a low-contrast grating with a randomly chosen orientation was presented (Fig 1iii) succeeded by a noise patch. The participants’ task was to adjust a response bar (Fig 1iv), such that it matched the orientation of the previously perceived probe stimulus. We focused on whether and how the preceding superposition grating and/or the two orthogonal gratings influenced the percept of the probe.
First we describe baseline performance for the probe stimulus alone, i.e., without preceding gratings (Fig 1, “Control Condition 1”). Overall, perceived orientations matched the presented orientations with small biases away from cardinal axes (maximum mean bias value 6.8° ± 6.2°SD over all participants, Fig 2A). Such biases  were reported to be particularly strong at intermediate stimulus contrasts  and were here consistently found across all conditions (Fig 2A/2B), including additional control conditions, where solely the superimposed gratings were shown with different timing (S1 File of S1 Fig).
(A) Control condition 1, in which only the probe stimulus was shown. (B) Test condition (superimposed gratings followed by gratings with a single orientation). Outlined panels (black box) depict data for conditions in which superimposed gratings were presented for 200 ms, the lower two panels show data for shorter (33 ms) presentation time of the superposition. Gray bars mark orientations where the probe stimulus was identical to the single grating orientation (see icons at left of each panel). In each plot red lines depict mean, colored dots represent a single participant’s judgments for a given orientation of the probe stimulus for each of 8 trials. Probe orientations oblique to the single gratings were measured twice and are separately depicted as +/-90°. Note that multiple individual adjustment responses were orthogonal to the probe stimulus orientations (= illusory percepts; see particularly the gray rectangles in runs with 200 ms of the superimposed gratings).
Surprisingly, in the test conditions, when the superimposed grating was followed by a single grating, a number of judgments clustered around orientations that were orthogonal to the to-be-matched probe orientation (Fig 2B, see “outliers” within gray bars and around nearby orientations). These “illusory percepts” of the difference between past and present orientations (which we refer to as perceptual errors) appeared in 7 of the 15 participants (dot colors in Fig 2 denote individual subjects) almost exclusively at probe orientations parallel to the single grating orientations present after the superposition grating (Fig 2, cf. gray bars; obtaining the illusory percept by chance in all 4 conditions: p<10−6, based on 106 shuffles of data between different orientations). This indicates that the effect may involve specific adaptation to the orientation , that had remained present after the switch from superimposed gratings. Importantly, the effect was also critically dependent on switch timing: when the superimposed gratings were shown for 200 ms (Fig 2B, outlined panel), the probe stimulus at the adapted orientation was 20 times (8.3% of trials) perceived as being orthogonal (or close to orthogonal) to its real orientation, whereas this was only 8 times (3.3% of trials) the case for the faster switch after only 33 ms (Fig 2B, two bottom panels). This is consistent with recent cat studies which showed that the generation of error signals representing the difference between past and present orientations needs time to build up and seem to be reliably formed when switch times are larger than 33 ms [15,27]. In contrast, over all control conditions only two errors (defined as judgments 4SD outside the mean of the density distribution obtained for control conditions) occurred within a total of 4680 trials (S1 File of S1 Fig). The significance of these observations was evaluated by resampling trials of each condition at the respective +45 and -45-degree probe orientations (data points within gray bars in Fig 2) and counting the number of such errors (i.e., illusory percepts) across the different conditions (Fig 3). Further we found that the density distribution of illusory percepts was similar to the distribution of judgments around the true orientation (Fig 4, 33 ms p = 0.360, 200 ms p = 0.668, see S1 File of S1 Text), pointing to a systematic effect rather than to accidental occurrence of erroneous responses. Such a similar scatter of orientation judgments may therefore reflect shared processing resources of signals encoding the difference between past and present and signals encoding the real current input. The fact that both signals are generated in a stimulus regime of high temporal dynamics and low contrast without higher-order expectations favors primary visual areas as their common cortical origin .
Illusory percepts were treated as errors, defined as judgments that deviated in orientation angle more than 4SD from the mean of the density distribution obtained for control conditions. For each condition, data for the +45° and -45° probe orientations were randomly resampled (106 iterations). (A) Error count distribution for the condition with superimposed gratings presented for 200 ms followed by grating with single orientation. Mean error count = 20 (red line). Probability to obtain 8 error counts (black line), p = 0.0015. (B) Same as (A) for the test condition where the switch from superposition to single orientations occurred after 33 milliseconds. Mean error count = 8 (black line). Probability to obtain 20 error counts (red line) or more, p = 0.00022. Probability to obtain 1 error count (green line) or less, p = 0.00026. (C) Across all participants, all trials, and all three control conditions (15 subjects, 8 trials, 3 conditions, 2 orientations of the probe stimulus (cf. gray areas in Fig 2), in total 720 cases) one error as defined above occurred (green line). Consequently, resampling of control conditions shows that the chance to obtain error counts in the range as found for the switch from superimposed to single gratings (see A and B) was zero. In all plots purple lines indicate 5th and 95th percentile.
(A) Illustration of judgments around true orientation (gray dots) and illusory percepts (i.e., errors, reported orientations orthogonal to the actually presented probe orientation, indicated with red dots) for both test orientations -45° and +45° (centered at 0°, n = 240 data points in each upper and lower panel) with fitted normal distributions (black curves). Vertical offsets of data points are to avoid overlap and are only for demonstration purposes. (B) Cumulative distribution functions (cdfs) of perceived errors (only cases within 4SD around 90° were used, see red dots in (A) within gray shaded areas; n = 18 for 200 ms, n = 7 for the 33 ms condition) and resamples of control conditions using the same number of cases per sample (18 and 7, respectively, see Methods). Red curves depict cdfs of errors. Gray line depicts mean and gray shading represents 5th and 95th percentile boundaries of resampled controls. All distributions were centered at 0°. Note that the cdf of all errors is within percentile boundaries of resampled controls except for one case in the 33 ms condition. Histogram insets show mean absolute distance of each resample from the non-resampled (“reference”) cdf of controls (yellow curve). Red line indicates the mean distance of errors from the reference cdf. Chance to obtain a resample with distance equal or greater than perception of errors: p = 0.668 for the 200 ms switch condition, p = 0.360 for the 33 ms switch condition. The graph shows that the error cases are not extreme values of a distribution of erroneous percepts (or percept variance) but represent a separate set of rare cases with illusory percepts orthogonal to the probe stimulus.
We revealed cases of illusory visual percepts in which perceived orientations represented the arithmetic difference between orientations seen in the immediate past and actually present orientations. The illusory percept cannot be explained by known tilt-aftereffects of adaptation [28,29] and its rare appearance is incompatible with predictive coding schemes in which error (i.e., difference) signals are “explained away”  during recurrent processing steps. Instead, our observations suggest that error signals do not always diminish but sometimes remain accessible at late processing stages forming awareness . The underlying neuronal difference signals may penetrate perception through post-perceptual decision, occasionally ignoring information about the actually present stimulus at higher hierarchy levels. Thus, our results may support the view that perception depends on late decision stages [31–34] at which multiple signals of recent stimulus history  are analyzed and integrated before reaching conscious awareness [35–38].
Interestingly, during free viewing of natural scenes intersaccadic durations in humans are on average 250 ms , thus, covering a similar time range as spanned by the superimposed gratings presented for 200 ms in our study. Note also that the illusory effect was less prevalent with shorter presentation time (33 ms) of the superposition. In fact, changes between low and high frequency stimulus regimes are proposed to allow for coarse-to-fine processing of visual information [40–48], providing an efficient strategy for the detection of salient structures in the environment . Thereby, longer stimulus durations (on the time scale of a single fixation) facilitate identification of orthogonal orientations , while reducing redundancy near adapting orientations . The rare appearance of the illusion in our study might reflect an occasional erratic overcalibration of such function.
Why is the illusion observed only occasionally? The exact causes of the observed rare cases of illusory perception of difference signals between past and present orientations remain open. Illusions are often revealed by artificial stimulus configurations that produce extremes of what the system has evolved to handle . In contrast to studies that evaluate effect significance by comparing test and controls through comparison of the variance introduced by the treatment (a varied factor, assumed to affect the whole sample/population in a similar way) with the combined variances within the groups, we here derived multiple errors as "outliers" that, however, cannot be explained by chance. In particular, using permutation tests, we evaluated that the incidence of perceiving errors was significantly different across all tested conditions and that errors (i.e., illusory percepts) show the same distribution as judgments estimating true orientations (Figs 3 and 4, respectively), thus, ruling out simple outlier explanations. Moreover, the fact that across all control conditions illusory percepts were “practically non-existent” clearly suggests that the illusion depends on the introduced dynamics by the stimulus switches from superposition to single gratings, and hence, on the availability of the resulting difference signal at higher levels . Our experiments in cats showed consistent generation of the difference signal in primary visual cortex with increasing switch time intervals . This is in accordance with the present results and is most likely based on increasing impact of inhibitory suppressive mechanisms over time [15,16,53]. Thus, the fact that in the present human study the illusory percept occurred very rarely may straightforwardly depend on currently unknown impact of factors producing difference signals (such as stimulus durations, contrast, spatial frequencies) that vary in individual subjects or on psychometric factors (e.g., internal noise, attention) that influence individual perceptual decision processes during which signals of current stimuli compete with simultaneously present difference signals, Fig 5). A systematic exploration of factors enhancing the present illusion is therefore needed in future studies to identify the exact causes of its appearance in individual subjects.
Presentation of the superposition of two gratings (stimulus past, shown bottom left) activated neurons in V1 that are tuned to the given orientations (red and blue peaks). If the superposition remains for more than ~30 ms, abrupt removal of one orientation facilitates a strong orientation-selective "off" response , while responses to the present (i.e., sustained) orientation undergo adaptive decrease (bottom right, cf. amplified red and reduced blue peak amplitudes, respectively; stippled black downward arrow indicates feedback). "Off" responses, or alternatively, the difference of the peaks’ amplitudes (Δ) are forwarded (straight red arrows) to error units ("Error", red framed circles, peaks at left represent their output) further downstream (Levelx). In parallel, representation units (“Repr”) carry adapted signals of the present orientation (blue peaks). Representation units at higher hierarchical levels (Levelx+1) receive bottom-up prediction errors (curved red arrow) and send predictions via top-down projections (dotted curved black arrow) that gradually downsize prediction errors (red peaks at left) in upstream and laterally connected (gray curved arrows) error units (for clarity only two hierarchical levels are depicted; see  for a detailed theoretical account and  for review). At a putative perceptual decision stage, signals from error units (top red peak) and representation units (top blue peak) compete with each other, potentially weighted and thresholded (stippled horizontal line), to finally form conscious perception. If signals conveyed by error units are of large amplitude (and not explained away, as in the case of left gray peaks), they might occasionally outplay signals from representation units (encoding the present stimulus). This gives rise to illusory perception of the difference between past and present stimulus (i.e., the removed orientation, top left). Note again that responses to the present orientation underlie V1 adaptation. In contrast to non-adapted responses (gray peaks at right), adaptively reduced amplitude in representation units (blue peaks at right) increases the chance of being outbalanced by error unit amplitudes (red peak at left).
Potential relationships of our results to a predictive coding scheme
At a more abstract level, given the framework of predictive coding models [1–4,18,19], we propose that the illusory percept may originate from an internally generated prediction error that, in our case, becomes occasionally conscious without higher-order expectations [22,30].
Interestingly, also in human EEG recordings, changes in stimulus orientation evoked a deviant signal that was most strongly tuned to orthogonal orientations . These signals were interpreted as correlates of a prediction error, coding the relative difference between predicted and observed orientations . Note that in the current study we used switches from superposition to single gratings. This allows interpretation of the “deviant” error signal as a result of computing the arithmetic difference of subsequent stimuli (i.e., as the difference between past and present orientations) rather than resulting from a complete stimulus exchange. Our previous experiments in cat primary visual cortex [15,16] showed that after the switch from superimposed to single gratings, populations of neurons encode the present (sustained, and thus, adapted) orientation with suppressed amplitudes, whereas populations tuned to the removed orientation produce a strong and persistent [54,55] “off” response , as depicted in Fig 5 (cf. changes of peak activity at bottom [past vs. present], blue and red, respectively). Signals following stimulus removal are commonly referred to as visual off-responses, which tend to increase with stimulus duration . Such off-responses are also found in the somatosensory cortex . Consequently, within the same cortical region two functionally distinct populations co-exist, one suppressed and one containing enhanced iconic memory [58,59], capable of providing different information within predictive coding circuits . “Off” responses may either directly signal “prediction errors” (i.e., reporting difference to past; Fig 5, left straight red arrow) or prediction errors may be read-out at subsequent stages (Fig 5, Levelx) by computing the relative difference of amplitudes signaled by both populations  (Fig 5, right straight red arrows).
As suggested in classical models of predictive coding [1,4], prediction errors may propagate via feedforward connections (Fig 5, curved red arrow) and are compared at higher hierarchical levels (Fig 5, “Levelx+1”, only one higher level is shown for clarity) that correct current estimates of the input signal and generate top-down prediction (Fig 5, black dotted curved arrow) to minimize residual errors (Fig 5, red and gray peaks at left sketch strong and weak residual errors, respectively). Finally, we hypothesize a perceptual decision process (Fig 5, top panel, horizontal stippled line illustrates perceptual threshold), which thresholds (or actively weights) input from residual error signals (peaks at left) and input from current stimulus signals (peaks at right). The occasional perceptual access to prediction errors may depend on internal noise of processes that produce perceptual decisions, possibly also involving attentional resources, or on computations responsible for generating (or sustaining) strong enough “off” responses (Fig 5, left red peaks). In subjects without illusory perception, the chosen stimulus contrast, the duration of the superposition stimulus, or other factors, may not be sufficient to produce the necessary strong “off” responses (left gray peaks) or fail to produce sufficient adaptation (right gray peaks). As seen in our data, short presentation times increase the tendency of encoding the current stimulus. Most likely because cortical "off" responses depend on suppressive processes  that need time to build up. Thus, in most cases signals encoding the current input (even though adapted) might be larger than signals carrying prediction errors (i.e., errors were sufficiently explained away, cf. left gray peaks).
Altogether, we present a most simple version of predictive coding, where persistent activity of previous input (i.e., prediction error-like signals) competes with adaptive response components of current input at fast time scales. The first coding step is based on our empirical results obtained in cat V1 , where we showed that removal of an orientation generates a tuned off response (Fig 5, red peak in bottom right pane), while responses to sustained orientations undergo adaptive decay (Fig 5, blue peak in bottom right panel). Accordingly, due to increased contribution of orientation-selective off-responses, the combination of adaptive- and off-response components results in a representation of the difference between the past and the present image, i.e., generating an error signal tuned to the removed orientation (Fig 5, cf. delta at bottom right). Error signals may then be reduced by feedback from higher processing levels (Fig 5, cf. peaks at left, Levelx and Levelsx+1), as suggested by predictive coding models of visual processing streams . Adapted signals tuned to the present orientation propagate in parallel up the hierarchy along representation units (Fig 5, cf. peaks at right). Finally, a perceptual decision is made by thresholding peak activity from error and representation units, resulting in a discrete decoding of either the present (Fig 5, top right) or the removed orientation (Fig 5, top left), where the latter signifies a difference (error) signal.
We present a visual illusion in which observers perceive the difference between previously seen and actual orientations rather than the actual orientation itself. The illusion was produced by presenting the superposition of two orthogonally oriented gratings followed by abrupt removal of one orientation after either 33 or 200 milliseconds. However, the illusory percept occurred only rarely. Thus, apart from a clear dependence of the effect on the stimulus time interval, other factors driving this illusion remain to be found.
As the illusion reflects the arithmetic difference between a previously seen and a current stimulus we interpret our finding within the framework of predictive coding. Predictive coding provides decrease in redundancy  particularly suited to low noise situations , but bears the risk of snap judgments in case of uncertain input dynamics. From a general modeling point of view the low number of illusory percept cases found here might therefore reflect a signature of an in-build instability that prevents the system from running into potentially wrong attractors in ambiguous stimulus situations [62–66]. As demonstrated in recent simulations that coupled predictive coding schemes to intrinsic brain dynamics, such a computational strategy ensures that conditional expectation maintains uncertainty, which in turn allows for flexible representations of contingencies and ambiguities in the world . As our experiments imposed neither specific demand on the task, nor provided additional contextual influence, further studies may unravel how the presently low frequency of the illusory percept varies with attention [11,12], with spontaneous activity , or with global changes in brain state . We here propose a speculative scheme of predictive coding, where error signals, emerging most probably from first steps of cortical processing dynamics, are not explained away but maintained and pushed through the perceptual hierarchy to potentially benefit flexible context-dependent and time-critical post-perceptual decisions [31,70–75] at higher cortical entities.
- 1. Rao RP, Ballard DH. Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nat Neurosci 1999;2(1):79–87. pmid:10195184
- 2. Ullman S. Sequence seeking and counter streams: a computational model for bidirectional information flow in the visual cortex. Cereb Cortex 1995;5(1):1–11. pmid:7719126
- 3. Mumford D. On the computational architecture of the neocortex. II. The role of cortico-cortical loops. Biol Cybern 1992;66(3):241–51. pmid:1540675
- 4. Friston K. A theory of cortical responses. Philos Trans R Soc Lond B Biol Sci 29-4-2005;360(1456):815–36. pmid:15937014
- 5. Grill-Spector K, Henson R, Martin A. Repetition and the brain: neural models of stimulus-specific effects. Trends Cogn Sci 2006;10(1):14–23. pmid:16321563
- 6. Murray SO, Kersten D, Olshausen BA, Schrater P, Woods DL. Shape perception reduces activity in human primary visual cortex. Proc Natl Acad Sci U S A 12-11-2002;99(23):15164–9. pmid:12417754
- 7. Summerfield C, Trittschuh EH, Monti JM, Mesulam MM, Egner T. Neural repetition suppression reflects fulfilled perceptual expectations. Nat Neurosci 2008;11(9):1004–6. pmid:19160497
- 8. Alink A, Schwiedrzik CM, Kohler A, Singer W, Muckli L. Stimulus predictability reduces responses in primary visual cortex. J Neurosci 2010;30:2960–6. pmid:20181593
- 9. Kok P, Jehee JF, de Lange FP. Less is more: expectation sharpens representations in the primary visual cortex. Neuron 26-7-2012;75(2):265–70. pmid:22841311
- 10. Ehinger BV, Konig P, Ossandon JP. Predictions of Visual Content across Eye Movements and Their Modulation by Inferred Information. J Neurosci 13-5-2015;35(19):7403–13. pmid:25972169
- 11. Gordon N, Tsuchiya N, Koenig-Robert R, Hohwy J. Expectation and attention increase the integration of top-down and bottom-up signals in perception through different pathways. PLoS Biol 2019;17(4):e3000233. pmid:31039146
- 12. Smout CA, Tang MF, Garrido MI, Mattingley JB. Attention promotes the neural encoding of prediction errors. PLoS Biol 2019;17(2):e2006812. pmid:30811381
- 13. Summerfield C, Egner T, Greene M, Koechlin E, Mangels J, Hirsch J. Predictive codes for forthcoming perception in the frontal cortex. Science 24-11-2006;314(5803):1311–4. pmid:17124325
- 14. Kaliukhovich DA, Vogels R. Stimulus repetition probability does not affect repetition suppression in macaque inferior temporal cortex. Cereb Cortex 2011;21(7):1547–58. pmid:21097993
- 15. Nortmann N, Rekauzke S, Onat S, Konig P, Jancke D. Primary visual cortex represents the difference between past and present. Cereb Cortex 2015;25(6):1427–40. pmid:24343889
- 16. Nortmann N, Rekauzke S, Azimi Z, Onat S, Konig P, Jancke D. Visual homeostatic processing in V1: when probability meets dynamics. Front Syst Neurosci 2015;9:6. pmid:25691861
- 17. Stefanics G, Kremlacek J, Czigler I. Visual mismatch negativity: a predictive coding view. Front Hum Neurosci 2014;8:666. pmid:25278859
- 18. Srinivasan MV, Laughlin SB, Dubs A. Predictive coding: a fresh view of inhibition in the retina. Proc R Soc Lond B Biol Sci 22-11-1982;216(1205):427–59. pmid:6129637
- 19. Dan Y, Atick JJ, Reid RC. Efficient coding of natural scenes in the lateral geniculate nucleus: experimental test of a computational theory. J Neurosci 15-5-1996;16(10):3351–62. pmid:8627371
- 20. Dong DW, Atick JJ. Temporal decorrelation: a theory of lagged and nonlagged responses in the lateral geniculate nucleus. Network Comput Neural Sys 1995;6(2):159–78.
- 21. Attneave F. Some informational aspects of visual perception. Psychol Rev 1954;61(3):183–93. pmid:13167245
- 22. Todorovic A, de Lange FP. Repetition suppression and expectation suppression are dissociable in time in early auditory evoked fields. J Neurosci 26-9-2012;32(39):13389–95. pmid:23015429
- 23. Bach M. The Freiburg visual acuity test—automatic measurement of visual acuity. Optom Vis Sci 1996;73(1):49–53. pmid:8867682
- 24. Jastrow J. On the judgement of angles and positions of lines. American Journal of Physiology 1892;5(2):214–48.
- 25. de Gardelle V, Kouider S, Sackur J. An oblique illusion modulated by visibility: non-monotonic sensory integration in orientation processing. J Vis 12-8-2010;10(10):6. pmid:20884471
- 26. Movshon JA, Lennie P. Pattern-selective adaptation in visual cortical neurones. Nature 26-4-1979;278(5707):850–2. pmid:440411
- 27. Benucci A, Ringach DL, Carandini M. Coding of stimulus sequences by population responses in visual cortex. Nature Neuroscience 2009;12:1317–24. pmid:19749748
- 28. Clifford CWG, Rhodes G. Fitting the mind to the world; adaptation and aftereffects in high-level vision. In: Clifford CWG, Rhodes G, editors. Advances in Visual Cognition. Oxford University Press; 2005.
- 29. Clifford CWG. Aftereffects. In: Goldstein EB, editor. Encyclopedia of Perception. Thousand Oaks, CA: Sage Publications Inc.; 2010. p. 13–6.
- 30. van Heusden E., Harris AM, Garrido MI, Hogendoorn H. Predictive coding of visual motion in both monocular and binocular human visual processing. J Vis 2-1-2019;19(1):3. pmid:30630191
- 31. Fritsche M, Mostert P, de Lange FP. Opposite effects of recent history on perception and decision. Curr Biol 20-2-2017;27(4):590–5. pmid:28162897
- 32. Kayser C, Shams L. Multisensory causal inference in the brain. PLoS Biol 2015;13(2):e1002075. pmid:25710476
- 33. Mamassian P. Visual confidence. Annu Rev Vis Sci 14-10-2016;2:459–81. pmid:28532359
- 34. Kornmeier J, Worner R, Bach M. Can I trust in what I see? EEG evidence for a cognitive evaluation of perceptual constructs. Psychophysiology 2016;53(10):1507–23. pmid:27387041
- 35. Herzog MH, Kammer T, Scharnowski F. Time Slices: What is the duration of a percept? PLoS Biol 2016;14(4):e1002433. pmid:27070777
- 36. Gregory RL. Perceptions as hypotheses. Philos Trans R Soc Lond B Biol Sci 8-7-1980;290(1038):181–97. pmid:6106237
- 37. Fahrenfort JJ, van LJ, Olivers CN, Hogendoorn H. Perceptual integration without conscious access. Proc Natl Acad Sci U S A 4-4-2017;114(14):3744–9. pmid:28325878
- 38. Dehaene S, Changeux JP. Experimental and theoretical approaches to conscious processing. Neuron 28-4-2011;70(2):200–27. pmid:21521609
- 39. Kuang X, Poletti M, Victor JD, Rucci M. Temporal encoding of spatial information during active visual fixation. Curr Biol 20-3-2012;22(6):510–4. pmid:22342751
- 40. Parker DM, Lishman JR, Hughes J. Temporal integration of spatially filtered visual images. Perception 1992;21(2):147–60. pmid:1513664
- 41. Jancke D. Orientation formed by a spot’s trajectory: a two-dimensional population approach in primary visual cortex. J Neurosci 15-7-2000;20(14):RC86. pmid:10875941
- 42. Ahissar E, Arieli A. Figuring space by time. Neuron 25-10-2001;32(2):185–201. pmid:11683990
- 43. Geisler WS, Albrecht DG, Crane AM, Stern L. Motion direction signals in the primary visual cortex of cat and monkey. Vis Neurosci 2001;18:501–16. pmid:11829297
- 44. Wörgötter F, Suder K, Zhao Y, Kerscher N, Eysel UT, Funke K. State-dependent receptive-field restructuring in the visual cortex. Nature 12-11-1998;396(6707):165–8. pmid:9823895
- 45. Desbordes G, Rucci M. A model of the dynamics of retinal activity during natural visual fixation. Vis Neurosci 2007;24(2):217–30. pmid:17640413
- 46. Rucci M, Iovin R, Poletti M, Santini F. Miniature eye movements enhance fine spatial detail. Nature 14-6-2007;447(7146):851–4. pmid:17568745
- 47. Ahissar E, Arieli A. Seeing via miniature eye movements: A dynamic hypothesis for vision. Front Comput Neurosci 2012;6:89. pmid:23162458
- 48. Meirovithz E, Ayzenshtat I, Werner-Reiss U, Shamir I, Slovin H. Spatiotemporal effects of microsaccades on population activity in the visual cortex of monkeys during fixation. Cereb Cortex 2012;22(2):294–307. pmid:21653284
- 49. Dragoi V, Sharma J, Miller EK, Sur M. Dynamics of neuronal sensitivity in visual cortex and local feature discrimination. Nat Neurosci 2002;5(9):883–91. pmid:12161755
- 50. Müller JR, Metha AB, Krauskopf J, Lennie P. Rapid adaptation in visual cortex to the structure of images. Science 27-8-1999;285(5432):1405–8. pmid:10464100
- 51. Eagleman DM. Visual illusions and neurobiology. Nat Rev Neurosci 2001;2(12):920–6. pmid:11733799
- 52. Vladimirskiy B, Urbanczik R, Senn W. Hierarchical Novelty-Familiarity Representation in the Visual System by Modular Predictive Coding. PLoS One 2015;10(12):e0144636. pmid:26670700
- 53. Pernberg J, Jirmann KU, Eysel UT. Structure and dynamics of receptive fields in the visual cortex of the cat (area 18) and the influence of GABAergic inhibition. Eur J Neurosci 1998;10(12):3596–606. pmid:9875339
- 54. Nikolic D, Hausler S, Singer W, Maass W. Distributed fading memory for stimulus properties in the primary visual cortex. PLoS Biol 2009;7(12):e1000260. pmid:20027205
- 55. Weichselgartner E, Sperling G. Continuous measurement of visible persistence. J Exp Psychol Hum Percept Perform 1985;11(6):711–25. pmid:2934504
- 56. Duysens J, Schaafsma SJ, Orban GA. Cortical off response tuning for stimulus duration. Vision Res 1996;36(20):3243–51. pmid:8944284
- 57. Kyriazi HT, Carvell GE, Simons DJ. OFF response transformations in the whisker/barrel system. J Neurophysiol 1994;72(1):392–401. pmid:7965022
- 58. Neisser U. Cognitive psychology. Englewood Cliffs, NY: Prentice Hall; 1967.
- 59. Di Lollo V. Temporal characteristics of iconic memory. Nature 19-5-1977;267(5608):241–3. pmid:865614
- 60. de Gardelle V, Waszczuk M, Egner T, Summerfield C. Concurrent repetition enhancement and suppression responses in extrastriate visual cortex. Cereb Cortex 2013;23(9):2235–44. pmid:22811008
- 61. Chalk M, Marre O, Tkacik G. Toward a unified theory of efficient, predictive, and sparse coding. Proc Natl Acad Sci U S A 2-1-2018;115(1):186–91. pmid:29259111
- 62. Girshick AR, Landy MS, Simoncelli EP. Cardinal rules: visual orientation perception reflects knowledge of environmental statistics. Nat Neurosci 2011;14(7):926–32. pmid:21642976
- 63. Tomassini A, Morgan MJ, Solomon JA. Orientation uncertainty reduces perceived obliquity. Vision Res 5-3-2010;50(5):541–7. pmid:20005889
- 64. Maloney RT, Clifford CW. Orientation anisotropies in human primary visual cortex depend on contrast. Neuroimage 1-10-2015;119:129–45. pmid:26093331
- 65. Sandamirskaya Y, Schöner G. An embodied account of serial order: how instabilities drive sequence generation. Neural Netw 2010;23(10):1164–79. pmid:20800989
- 66. Weilnhammer V, Stuke H, Hesselmann G, Sterzer P, Schmack K. A predictive coding account of bistable perception—a model-based fMRI study. PLoS Comput Biol 2017;13(5):e1005536. pmid:28505152
- 67. Friston K, Breakspear M, Deco G. Perception and self-organized instability. Front Comput Neurosci 2012;6:44. pmid:22783185
- 68. Hesselmann G, Kell CA, Eger E, Kleinschmidt A. Spontaneous local variations in ongoing neural activity bias perceptual decisions. Proc Natl Acad Sci U S A 5-8-2008;105(31):10984–9. pmid:18664576
- 69. Shen L, Han B, Chen L, Chen Q. Perceptual inference employs intrinsic alpha frequency to resolve perceptual ambiguity. PLoS Biol 2019;17(3):e3000025. pmid:30865621
- 70. Heekeren HR, Marrett S, Ungerleider LG. The neural systems that mediate human perceptual decision making. Nat Rev Neurosci 2008;9(6):467–79. pmid:18464792
- 71. Cisek P, Pastor-Bernier A. On the challenges and mechanisms of embodied decisions. Philos Trans R Soc Lond B Biol Sci 5-11-2014;369(1655).
- 72. Lepora NF, Pezzulo G. Embodied choice: how action influences perceptual decision making. PLoS Comput Biol 2015;11(4):e1004110. pmid:25849349
- 73. Rahnev D, Nee DE, Riddle J, Larson AS, D’Esposito M. Causal evidence for frontal cortex organization for perceptual decision making. Proc Natl Acad Sci U S A 24-5-2016;113(21):6059–64. pmid:27162349
- 74. Sterzer P. Moving forward in perceptual decision making. Proc Natl Acad Sci U S A 24-5-2016;113(21):5771–3. pmid:27179006
- 75. Heeley DW, Buchanan-Smith HM. Orientation acuity estimated with simultaneous and successive procedures. Spat Vis 1992;6(1):1–10. pmid:1536826