Temporal dynamics of task-irrelevant visual perceptual learning with threshold and suprathreshold stimulus exposure: a pilot study
Samuel Rasche
Principal investigators: dr. S.M. Frank & dr. T. Watanabe Laboratory for Cognitive and Perceptual Learning
Brown University
Abstract
Frequent exposure to visual stimuli may result in long-term performance improvement on a visual task. This is phenomenon is known as visual perceptual learning (VPL). In addition, unattended task-irrelevant features may be learned as well when these are presented in conjunction with task-relevant features. This phenomenon is known as task-irrelevant perceptual learning (TIPL). Importantly, previous studies have shown that TIPL only occurs with exposure to weakly visible (subthreshold or threshold level) irrelevant stimuli and not with salient (suprathreshold level) irrelevant stimuli. The current pilot study aims to replicate these previous findings in order to investigate the neural correlates of TIPL in a future study. The experiment took place at a psychophysics laboratory or remotely on personal desktops. Subjects (N = 9) completed 12 sessions of a Rapid Serial Visual Presentation (RSVP) task that was surrounded by a dynamic random dot (DRD) display moving coherently into a direction at either threshold level or suprathreshold level of visibility. The coherent motion was in the visual periphery and task-irrelevant. Results showed no significant improvement in direction discrimination at either coherent motion level and thus are the results insufficient to support a task-irrelevant learning effect. Nevertheless, a large effect size and the expected trend of improved performance with exposure at threshold level and no improvement with exposure at suprathreshold level was observed after 12 training sessions. Interestingly, this trend was not observed after 6 training sessions, indicating that many training sessions are needed for TIPL to occur. Moreover, these results were found irrespective of experimental environment (that is, at a psychophysics laboratory or at-home). In conclusion, the current experimental design showed promising results in detecting a TIPL effect and may therefore be considered in subsequent neuroimaging studies.
Introduction
The brain has to learn features from the environment that are relevant to attain the goals of the organism. In order to learn novel information, the brain is able to adapt to frequently encountered stimuli and become increasingly sensitive to these specific stimuli. At the same time, it needs to retain previous knowledge and therefore needs to maintain a degree of stability. Hence, not all novel information it is exposed to can be learned. This is known as the plasticity-stability dilemma (Grossberg, 1980; Mermillod et al., 2013). Higher cognitive functions deploy strategies to select relevant information for learning and filter out irrelevant information.
The increased sensitivity by repeated exposure to visual stimuli is known as visual perceptual learning (VPL) (Lu et al., 2011; Sasaki et al., 2010; Shibata et al., 2011). This long-term enhanced performance as a result of visual experience is for example illustrated in object recognition (Furmanski & Engel, 2000), reading (Chung et al., 2004) and the
identification of tumors on X-rays (Sasaki et al., 2010). The enhancement is reflected in the brain by reshaped (perceptual) systems, which is known as neural plasticity. Neural plasticity as a consequence of perceptual learning has been shown by neurophysiological studies (Kourtzi et al., 2005; Law & Gold, 2008; Recanzone et al., 1993; Yang & Maunsell, 2004). In addition, behavioral studies have used specificity of learning to the trained position as indicative for neural plasticity (Gilbert et al., 2001; Karni & Sagi, 1991; Seitz et al., 2009).
VPL has been related to plasticity in early visual areas (Frank et al., 2018; Furmanski et al., 2004; Karni & Sagi, 1991; Maertens & Pollmann, 2005; Schoups et al., 2001; Schwartz et al., 2002; Yotsumoto et al., 2014; Yotsumoto et al., 2008), but also to later stages of
processing (e.g. the parietal cortex) (Ahissar & Hochstein, 1997; Gilbert et al., 2001; Kourtzi et al., 2005; Law & Gold, 2008; Sigman et al., 2005). In general, perceptual learning will occur for stimuli that are relevant to attain the goals of the individual (Sasaki et al., 2010;
Seitz & Watanabe, 2009). Attention plays a crucial role in this; various studies have shown that only attended features are learned (Ahissar & Hochstein, 1993; Schoups et al., 2001; Shiu & Pashler, 1992). Attention enhances the task-relevant features, while inhibiting the irrelevant features (Sasaki et al., 2010).
VPL is predominantly specific to the visual feature (e.g. an orientation) that has been presented during the training (Karni & Sagi, 1991; Sasaki et al., 2010; Shibata et al., 2012). Various studies have demonstrated that the ability to discriminate between motion directions can be improved by means of training and this has been related to plasticity in visual areas (e.g., Ball & Sekuler, 1987; Frank et al., 2018; Shibata et al., 2012). Interestingly, Watanabe et al. (2001) demonstrated that exposure to an irrelevant, invisible background motion moving into a certain direction caused improvement in discriminating that specific motion direction in subsequent testing, indicating that the unattended, invisible motion was learned. Thus, contrary to earlier beliefs, this finding demonstrated that sensory plasticity may arise with frequent exposure in the absence of attention and awareness. VPL does therefore not only occur for attentively processed task-relevant stimuli (Ahissar & Hochstein, 1993; Schoups et al., 2001; Shiu & Pashler, 1992) but also for task-irrelevant stimuli (Seitz & Watanabe, 2005).
How can this task-irrelevant perceptual learning (TIPL) effect be explained? Only the motion direction that was correlated with the target was learned, indicating that irrelevant stimuli should coincide with relevant stimuli (Seitz & Watanabe, 2003). In addition, Seitz et al. (2009) demonstrated that pairing an unperceived stimulus with a reward is sufficient for TIPL to occur. Thus, purely passive exposure cannot account for the effect. Rather, TIPL seems to be driven by reinforcement signals that are triggered by stimuli that are paired with targets (which serve as internal rewards) or by external rewards (Sasaki et al., 2010; Seitz & Watanabe, 2009; Watanabe & Sasaki, 2015). Thus, in contrast to attentional mechanisms,
reward enhances perceptual learning irrespective of whether the feature is task-relevant or irrelevant (Kim et al., 2015; Sasaki et al., 2010; Watanabe & Sasaki, 2015).
TIPL seems mostly to be related to plasticity in early visual areas (Pilly et al., 2010; Seitz et al., 2009; Watanabe et al., 2002). For instance, TIPL occurred only for the local motion of the moving dots, which is processed in low-level areas, in contrast to the global coherent motion, which is processed at higher stages in the brain (Watanabe et al., 2002). Moreover, Seitz et al. (2009) demonstrated that unconscious perceptual learning was specific to the eye to which the stimuli were presented, which is indicative for early stages of visual processing (although some degree of ocular specificity extends to higher visual areas (Seitz & Watanabe, 2009)). Thus, these findings suggest that plasticity in low-level visual areas are related to TIPL.
Nevertheless, the role of higher-level processes in TIPL is significant. Namely, higher-level processes gate whether TIPL occurs. As mentioned above, attentional
mechanisms suppress task-irrelevant features (Choi et al., 2009; Sasaki et al., 2010; Shiu & Pashler, 1992). Moreover, TIPL seems mostly to occur for weak (subthreshold or threshold) motion coherence levels and does not occur for salient (suprathreshold) coherence levels, at least not in younger individuals (Chang et al., 2014; Tsushima et al., 2008). Performance may even become worse with suprathreshold level exposure (Chang et al., 2014; Tsushima et al., 2008). It has therefore been proposed that the learning of irrelevant (sub)threshold stimuli may be due to lack of inhibitory control. That is, the weak signal may not be detected by the attentional or executive control systems and is therefore not suppressed (Ahissar &
Hochstein, 2004; Chang et al., 2014; Tsushima et al., 2006; Tsushima et al., 2008). The influence of executive areas in filtering out irrelevant stimuli was elegantly shown by Tsushima et al. (2006). Using functional magnetic resonance imaging (fMRI), they found decreased blood oxygenation level dependent (BOLD) activity in the medial temporal cortex
(MT+, an extrastriate cortical area that is activated by motion perception (Born & Bradley, 2005; Tootell et al., 1995)) with suprathreshold coherence motion exposure, whereas MT+ BOLD activity was increased with subthreshold motion exposure. At the same time, the lateral prefrontal cortex (LPFC) was activated with suprathreshold motion exposure, but not with subthreshold motion exposure. This may indicate that the visual areas were suppressed by control areas when exposed to irrelevant suprathreshold motion, thereby preventing the processing of this irrelevant stimulus.
In summary, reinforcement signals enhance visual features irrespective of their relevance. With lack of inhibition by higher cognitive functions, irrelevant features may be learned. Although TIPL may be gated by higher-level processes, plasticity in low-level visual areas seems to reflect a learning effect. However, the specific neural and neurochemical mechanisms involved are not yet clear. With regard to VPL, early visual area activation increased during the initial stages of training (Schwartz et al., 2002; Yotsumoto et al., 2008). Whether this also holds for TIPL remains yet to be solved. In addition, it is not clear whether higher-level processes will exert more influence over suprathreshold distractors during the course of learning.
The current and future studies will aim to answer these questions. In order to do so, subjects will be exposed to task-irrelevant motions at threshold and suprathreshold coherence levels. The threshold will be determined for each subject from a psychometric function that is based on the response accuracy with different motion coherence levels (see Chang et al. 2014). The 80% threshold will be chosen as the individual threshold. The suprathreshold level is determined by multiplying the threshold by four. In order to learn, 12 training sessions will be conducted on separate days in which the subjects perform a Rapid Serial Visual Presentation (RSVP) task while being exposed to the motion in the background, rendering the motion task-irrelevant. To map the neurochemical mechanisms, three training
sessions will be conducted inside the MRI scanner. Univariate fMRI BOLD imaging will be used to measure activity in early visual areas and prefrontal areas. To further investigate the suppression hypothesis, magnetic resonance spectroscopy (MRS) will be used. This brain imaging method allows quantification of the concentration of neurochemicals in a specified brain area. Plasticity has been related to the ratio of excitatory (glutamate) to inhibitory (GABA) neurotransmitters in the brain regions that are involved in the learning of that particular skill (Bang et al., 2018; Frangou et al., 2018; Shibata et al., 2017). For example, it has been demonstrated that the degree of motor learning correlated positively with a decrease of GABA in the primary motor cortex (Floyer-Lea et al., 2006; Kim et al., 2014; Stagg et al., 2011), suggesting that decreases in local GABA concentration facilitate plasticity in that area (Frangou et al., 2018). In addition, increased glutamate correlated positively with plasticity (Bang et al., 2018; Cohen Kadosh et al., 2015; Nikolova et al., 2017). Therefore, an increased ratio of glutamate/glutamine (Glx) to GABA can be considered to be indicative of learning and plasticity.
In line with previous studies (Chang et al., 2014; Tsushima et al., 2008), it is expected that learning will occur with exposure to threshold coherent motion, whereas no learning will occur with exposure to suprathreshold coherent motion. Similar to task-relevant VPL
(Furmanski et al., 2004; Schwartz et al., 2002; Yotsumoto et al., 2008), it is expected that the BOLD activity in V1 and the MT+ will increase with threshold motion coherence exposure, thus when learning occurs. In addition, it is expected that this learning will be reflected by an increased ratio of Glx to GABA (E/I ratio) in these areas, to suggest plasticity. In contrast, with suprathreshold exposure it is expected that the dorsolateral prefrontal cortex (dlPFC) will increase in BOLD activity, to reflect the suppression of the irrelevant stimuli and thereby preventing learning. Likewise, it is expected that the E/I ratio in V1 and MT+ will decrease, to indicate the inhibition by the dlPFC and thereby non-learning (Figure 1).
Figure 1. Schematic representation of the brain and the expected interactions. Areas that increase in BOLD activity are denoted in black. With exposure at threshold level, an increase in BOLD activity and E/I ratio is expected in V1/MT+, while the dlPFC is not activated. With exposure at suprathreshold level, the early visual cortex activates the dlPFC, which in turn suppresses V1/MT+. This is reflected by increased BOLD activity in the dlPFC and a decrease in E/I ratio in V1/MT+.
With regard to the behavioral part of the experiment, certain predictions can be made as well. In line with previous studies, it is expected that the ratio of signal dots to the total number of dots can be correctly identified in 80% of cases when this ratio is 15% or higher. That is, the threshold for coherent motion detection is expected to be around 15% on average with the specific visual setup and stimuli used for this study (Chang et al., 2014; Tsushima et al., 2008). In addition, the ability to discriminate the direction in which part of the dots move coherently is expected to improve during the course of training (compared to baseline) with exposure at threshold level, but not with exposure at suprathreshold level. This improvement with threshold exposure may already be present after six training sessions, but is expected to be more salient at the end of the training (12 sessions) (Chang et al., 2014; Tsushima et al., 2008). At last, following the results of a previous study (Frank et al., under revision), accuracy on the RSVP task is expected to be around 60% on the first training session, peaking (90% or higher accuracy) at the third session, and then plateau. Since the coherent background motion is task-irrelevant, no performance difference between the two conditions is expected on the RSVP task.
Due to the COVID-19 crisis, the brain imaging part of the study will be omitted in the current study. The purpose of this study is to replicate previous behavioral results and to determine the time-course of the development of threshold and suprathreshold TIPL. If the expected results are found, the current experimental design may be used in future
neuroimaging studies.
Method
Subjects
To ascertain that TIPL occurs with the considered design, a behavioral pilot study was conducted with healthy young subjects (N = 9, 3 male). The average age was 23.9 (SD = 3.7) years and all subjects were right-handed and had normal or corrected-to-normal vision. The subjects were recruited via Brown University fliers, which were distributed throughout the campus (before the COVID-19 crisis) or via e-mail (during the COVID-19 crisis).
Procedure
The first four subjects completed the experiment in a psychophysics laboratory at Brown University, the remaining five subjects completed the experiment remotely on personal desktops. The experiment began with an informed consent form approved by the Brown University Institutional Review Board (both for the in-person and remote version of the study), an eligibility checklist, and a voluntary demographic questionnaire. After filling in the questionnaires, the subject moved on to the experimental tasks. The task and stimuli were programmed and generated in Matlab (version 2019b; The MathWorks, Natick, MA), using the Psychophysics Toolbox (PTB version 3.0.16; Brainard, 1997; Pelli, 1997). The first four subjects that completed the experiment at Brown University performed the experimental tasks in a dark room, presented on a 24-inch computer. Gamma-correction was performed to
adjust the luminance. The viewing distance was 60 cm, which was maintained by having the participant lean on a chinrest.
Due to the COVID-19 crisis, the data of the remaining five subjects were collected remotely. This was done by sending the files of the task to subjects that had Matlab (version 2019b, The MathWorks, Natick, MA) and Psychophysics Toolbox (PTB version 3.0.16; Brainard, 1997; Pelli, 1997) installed on their personal computers. The procedure remained the same, that is, the same approved questionnaires were filled in and the same tasks were performed. However, it was not possible to perform a gamma-correction on the various computers, nor could it be ascertained that the subjects were maintaining the specified distance. Nevertheless, the subjects were encouraged to maintain a distance of 60 cm and perform the sessions in a dark room, thereby mimicking the original experimental setting as closely as possible. The experiment consisted of 15 sessions, each session carried out on a different day (Figure 2). Three of the 15 sessions were test sessions, the other 12
training/exposure sessions. Based on previous studies (Chang et al., 2014; Tsushima et al., 2008) it was expected that about 12 training sessions would be necessary to find a difference between conditions, but to explore the temporal dynamics of TIPL a test session after six sessions was included. There were two conditions, a threshold and a suprathreshold condition. The subjects were assigned randomly to either condition. Four subjects were assigned to the threshold condition, the other five subjects to the suprathreshold condition. Of the subjects in the threshold condition, two completed the experiment at the psychophysics laboratory, the other two subjects completed the experiment remotely. Of the subjects in the suprathreshold condition, two completed the experiment at the psychophysics laboratory, the other three subjects completed the experiment remotely.
Figure 2. Experimental design. The study included a total of 15 sessions. The experiment started with a test session on day one, followed up by six training sessions. To assess the course of learning an intermediate test session was conducted on session eight, again followed up by six training sessions. The final (15th) session was
another test session to assess the final performance. Only a maximum of one session was conducted on a single day. The 15 sessions were completed in 25 days on average (SD = 6.0). Each session lasted about 40 minutes.
Three different tasks were completed in session number one. First, the visibility threshold of motion coherence was established for each participant. The participants viewed a dynamic random dot (DRD) display (Figure 3a). In this display, either all the dots were moving at random, or part of the dots were moving coherently into a randomly chosen direction (10°, 70°, 130°, 190°, 250°, and 310°), while the rest of the dots moved randomly. The percentage of coherently moving dots was 3, 13, 23, 33 or 53% and was randomly determined from trial to trial. The task included a total of 300 trials and the dots were presented for 500 ms per trial. Half of the trials included coherently moving dots at the various percentages and the other half of trials included completely random moving dots. To determine the threshold for motion coherence visibility, the participant was asked to judge after each trial whether the dots moved coherently or randomly (two-alternative forced choice task). The participants responded by pressing one of two buttons on the keyboard
corresponding to yes or no for coherent motion. No feedback about response accuracy was provided. The threshold was determined by performing a psychometric fit (logistic
regression) across the response accuracy with different motion coherence levels (3, 13, 23, 33 and 53%). The 80% threshold was chosen as the individual threshold (Chang et al., 2014). After completing this 15-minute long task, a motion direction task was executed (Figure 3b). This task was similar to the previous one, but this time participants indicated in which
direction the dots were moving by clicking an arrow that best represented the direction. Coherent motion was presented on every trial, thus no trials with only randomly moving dots were included. The coherence levels varied randomly between 0.3, 0.6, 1.0 and 4.0 times the individual threshold (which was established in the first task). Again, no feedback about response accuracy was provided. The direction discrimination task included 480 trials and took about 30 minutes to complete.
a. b.
Figure 3. Tasks performed in the test session. a. Coherence threshold determination task. A DRD display was presented with either randomly moving dots or a number of dots (3, 13, 23, 33 or 53%) moving coherently into a certain direction (10°, 70°, 130°, 190°, 250°, and 310°). The direction and number of dots moving coherently varied per trial. Subjects indicated whether the dots moved randomly or whether they observed coherent motion (irrespective of the direction). b. Direction discrimination task. After determining the threshold, the subjects’ ability to discriminate between the directions was assessed. Again, a DRD display was presented with a number of dots (varying per trial) moving coherently into a certain direction. The subject was asked to indicate in which direction the dots were moving. Coherent motion was presented on every trial. The coherence levels were varied between 0.3, 0.6, 1.0 and 4.0 x the individual threshold.
The first test session was followed up by six exposure sessions. In these exposure sessions there were two conditions: threshold and suprathreshold exposure. In the threshold condition, subjects were exposed to their predetermined threshold for motion coherence visibility. In the suprathreshold condition the subjects were exposed to their threshold multiplied by four (Chang et al., 2014). The subject was only exposed to one direction, as in the paradigm of Watanabe et al. (2001). The motion direction was randomly determined for each subject prior to the experiment. This approach was adopted instead of the paradigm of Seitz & Watanabe (2003), wherein the subject is exposed to various directions. The reason
for this was that the current study is interested in the difference in learning between threshold and suprathreshold exposure, and not necessarily in which direction is learned. No task was imposed on the DRD displays. The DRD display surrounded a RSVP task that engaged the subject’s attention, making the motion exposure task-irrelevant. In the RSVP task, a sequence
was presented of eight randomly chosen blurred images, two of which were targets. The target images were animals (red bird, yellow fish, green butterfly, blue bug). The distractor images were different vegetables, fruits and flowering plants. There were a total of eight different images for each of the distractor categories. A total of 150 trials were included and these sessions took about 30-40 minutes to complete. Each trial was four seconds long and each image was presented for a total of 350 ms, with 75 ms on and off ramps before and after each image. A central fixation cross was presented during the ramps. All RSVP images were overlaid with a mask of randomly colored pixels in red, green, yellow and blue that replaced 81% of the original pixels in each image. This was done to increase RSVP task difficulty, following pilot results demonstrating that with this setup initial RSVP task
performance is around 60-70% correct across subjects. The subjects were given four response options (corresponding to the four animals) and were asked to indicate after each trial which two animals were presented on that given trial, and in which order. Feedback on accuracy was provided after each trial. After six exposure sessions, a test session was conducted again wherein the two tasks of the first session (motion detection and motion discrimination) were completed again, in order to determine whether learning had occurred and to check whether the threshold had shifted. This was followed up by another six exposure sessions. At last, a final test session was conducted wherein the two initial tasks were completed, as in the first and second test sessions.
Data analysis
As mentioned above, the threshold for coherent motion detection was calculated by
performing a psychometric fit across the response accuracy with different motion coherence levels (3, 13, 23, 33 and 53%). The 80% threshold was chosen as the individual threshold (Chang et al. 2014). The average and standard error to the mean (SEM) were calculated. In addition, independent t-tests were conducted to check for any differences in coherence threshold level between conditions and between experimental settings. Moreover, a mixed-design ANOVA was conducted with test session (pre, intermediate and post) as within-subject factor and coherent motion level (threshold or suprathreshold) as between-within-subject factor to see whether the threshold for coherent motion visibility changed over time and if this interacted with condition. The assumption of sphericity was checked with Mauchly’s test. Any significant effects were followed-up with independent t-test. The various t-tests were accompanied with an effect size (Cohen’s d) calculation.
For each RSVP training stage, the accuracy in percentage and the SEM was
calculated. Only trials where subjects correctly identified the two target animals in the correct order were scored as correct. A mixed-design ANOVA was conducted with training session (1-12) as within-subject factor and coherent motion level (threshold or suprathreshold) as between-subject factor to check whether accuracy improved on the RSVP task over time and if this interacted with condition.
With regard to the direction discrimination, performance response accuracy in percentage was calculated for each coherence level (i.e., 0.3, 0.6, 1.0 and 4.0 x threshold) in each test stage for the exposed motion direction. Next, the difference between pre-test and intermediate test (mid-pre) and the difference between pre-test and post-test (post-pre) was taken by subtracting the percentage of correct responses for each coherence level in the pre-test from the intermediate pre-test and the post-pre-test, respectively, and by summing across
coherence levels in order to calculate the improvement in percentage. This was done only for the direction the subject was exposed to in the training sessions, because previous studies have shown that it is this direction the subject becomes better at discriminating (Seitz & Watanabe, 2003). The SEM was calculated as well. One-sample t-tests were used to see whether performance improved in the two conditions. In addition, independent t-tests were used to check for any differences between conditions. The effect size (Cohen’s d) was calculated. Moreover, a mixed-design ANOVA was conducted as well with test session (pre, intermediate and post) as within-subject factor and coherent motion level (threshold or suprathreshold) as between-subject factor to see whether direction discrimination improved and if this interacted with condition. The assumption of sphericity was checked with
Mauchly’s test. The effect size measure partial eta squared (partial η²) was calculated
alongside with the mixed-design ANOVA.
Results
Coherence Threshold
In accordance with previous studies (Chang et al., 2014; Tsushima et al., 2008), the average coherence threshold for motion visibility over all subjects was 16.4 ± 1.3% prior to training. In order to check for a practice effect in motion detection, the coherence threshold was examined on each test session. A decrease in threshold level is observed over time with exposure, resulting in a shift in average coherence threshold to 14.7 ± 1.5% after six training sessions and to 11.7 ± 1.6% after the 12 training sessions. Further looking at the different conditions, the average threshold in the threshold condition was 15.0 ± 1.8% prior to training, 11.8 ± 2.0% during training and 12.0 ± 2.1% after training, whereas the average threshold in the suprathreshold condition was 17.5 ± 1.8% prior to training, 17.1 ± 2.0% during training and 11.5 ± 2.6% after training. The two conditions did not differ significantly in coherence
threshold prior to training (t(7) = -0.97, p = 0.36, d = -0.65), nor during training (t(7) = -2.19, p = 0.06, d = -1.47), nor after training (t(7) = 0.12, p = 0.90, d = 0.09). A mixed-design ANOVA was conducted to see whether the decrease in coherent motion threshold was significant and interacted with coherent motion level. Mauchly’s test indicated that the assumption of sphericity had not been violated (χ²(2) = 2.27, p = 0.32). The analysis revealed that the decrease in coherent motion threshold over time was significant (F(2,14) = 7.17, p = 0.007), but did not interact with condition (F(2,14) = 2.90, p = 0.09). Independent t-test revealed a significant difference in coherence threshold between pre-test and post-test (t(16) = 2.29, p = 0.04, d = 1.08), but no difference between pre-test and intermediate test(t(16) = 0.85, p = 0.41, d = 0.40), nor a difference between intermediate test and post-test (t(16) = 1.39, p = 0.18, d = 0.66). Taking together, these results indicate that the threshold for coherent motion visibility had decreased after training, irrespective of the exposed coherent motion level. The change in coherent motion threshold level after training has also been observed by Tsushima et al. (2008).
To examine whether experimental settings influence the coherence threshold level, a distinction has also been made between the psychophysics laboratory at Brown University and the remote experiment on personal desktops. The average threshold in the psychophysics laboratory was 12.7 ± 0.6% prior to training, 11.6 ± 2.1% during training and 9.1 ± 2.7% after training, whereas the average threshold for remote subjects was 19.4 ± 0.8% prior to training, 17.2 ± 1.3% during training and 13.8 ± 1.5% after training. The two experimental settings differed significantly in coherence threshold prior to training (t(7) = -6.58, p < 0.001, d = 4.42), during training (t(7) = 2.37, p = 0.0495, d = 1.59), but not after training (t(7) = -1.58, p = 0.16, d = -1.06). These results indicate that prior to training a lower threshold for coherent is obtained in the psychophysics laboratory compared to personal desktops. However, this difference had vanished after training. The initial increased visibility in the
psychophysics laboratory may be due to the stricter setting and presumably better quality of the desktop compared to a personal desktop.
Rapid Serial Visual Presentation
As seen in Figure 4, the initial accuracy on the RSVP task was at 77.0 ± 4.7% and then an increase in performance is observed, with performance plateauing (97.0 ± 1.2%) around the third training session. Two subjects were considered outliers on the RSVP task and therefore their data was not included. A mixed-design ANOVA revealed a significant change in performance on the task (F(11,55) = 15.11, p = 0.001) and this was constant across coherent motion level (F(11,55) = 3.34, p = 0.08), indicating that performance on the RSVP task improved in either condition.
Figure 4. The average ± SEM performance on the Rapid Serial Visual Presentation (RSVP) task across subjects (N = 7). The x-axis represents the training sessions. The y-axis represents the percentage correct.
Direction Discrimination
As shown in Figure 5, an average improvement in performance of 55 ± 27% on the direction discrimination task was observed in the threshold condition (that is, ~15% coherent motion) after training. This improvement was however not significantly different from zero (t(3) =
2.04, p = 0.13), indicating that no learning occurred. Nevertheless, a large effect size is observed (d = 1.02) and only one subject out of the four did not show net improvement. A post-hoc power analysis, executed in G*Power 3.1 (Faul et al., 2007), indicated that this sample size with an effect size of d = 1.02 would lead to a power of 0.30 in detecting a difference from zero. In the suprathreshold condition (that is, ~70% coherent motion) there was a slight decrease in performance on the direction discrimination task in the final test session. The average performance was 6 ± 19.5%, which did not differ from zero (t(4) = -0.31, p = 0.77, d = 0.14), indicating that learning did not occur with exposure to
suprathreshold motion coherence. A post-hoc power analysis indicated that this sample size with an effect size of d = 0.14 would lead to a power of 0.06 in detecting a difference between from zero. When comparing the two conditions after training, again a large effect size but no significant difference was found (t(7) = 1.88, p = 0.10, d = 1.24). Here a post-hoc power analysis indicated that this sample size with an effect size of d = 1.24 would lead to a power of 0.36 in detecting a difference between the two conditions. Further, a mixed-design ANOVA was conducted. Mauchly’s test indicated that the assumption of sphericity had not been violated (χ²(2) = 2.61, p = 0.27). Neither the main factor of test was significant (F(2,14) = 0.97, p = 0.40, partial η² = 0.12), nor the interaction between test and coherent motion level (F(2,14) = 1.76, p = 0.21, partial η² = 0.20), indicating that performance did not change and this was constant across conditions. In conclusion, no significant effect was found in the threshold condition and thus it cannot be concluded that learning occurred. However, the effect sizes are large and the observed trend of increased performance in the threshold condition but not in the suprathreshold condition after training is in accordance with the expectations and previous studies (Chang et al., 2014; Tsushima et al., 2008).
Interestingly, the expected trend is not yet observed at the intermediate test (Figure 5). An average of improvement of 20 ± 37.5% in the threshold condition did not differ
significantly from zero (t(3) = 0.53, p = 0.63, d = 0.27). Similar improvement was found in the suprathreshold condition (16 ± 28.2%), which did not differ from zero either (t(4) = 0.57, p = 0.60, d = 0.25), indicating that no learning occurred after six training sessions.
When looking at the two different experimental settings separately, no difference between the laboratory setting and the remote setting was observed. On average, both settings showed similar net improvement in the threshold condition (laboratory: 57.5 ± 57.5%,
remote: 52.5 ± 32.5%), and no net improvement in the suprathreshold condition (laboratory: 10 ± 45%, remote: -16.7 ± 21.3%).
Figure 5. The average ± SEM performance improvement in direction discrimination for the threshold condition and suprathreshold condition after 6 (intermediate test, blue) and 12 (post-test, red) training sessions. The x-axis represents the exposed motion coherence level for the intermediate test (left, blue) and final test (right, red). The y-axis represents the improvement in percentage which is calculated by subtracting the performance on the pre-test from the intermediate pre-test and pre-pre-test from post-pre-test.
Discussion
In the present study, TIPL was investigated. More specifically, the influence of strong versus weak irrelevant stimuli was examined. While performing an RSVP task, human subjects (N = 9) were exposed to coherent motion at either threshold level (perceptually weak) or
surrounding the RSVP task, rendering the motion task-irrelevant. No significant improvement on direction discrimination was found after 6 or 12 training sessions for either the subjects exposed to coherent motion at threshold level (N = 4) or the subjects exposed at
suprathreshold level (N = 5), indicating that learning of task-irrelevant motion did not occur. However, the expected trend of improvement with threshold level exposure but no
improvement with suprathreshold exposure was observed, along with large effect sizes. How could improvement on direction discrimination with threshold level exposure occur, but not with suprathreshold level exposure? Previous research suggests that higher level brain regions may not detect the weak bottom-up irrelevant motion and therefore fail to suppress the irrelevant motion (Ahissar & Hochstein, 2004; Chang et al., 2014; Tsushima et al., 2006; Tsushima et al., 2008). Consequently, the failure in evoking top-down attentional and control mechanisms results in enhanced sensitivity to the external environment and hence improved novelty detection. In contrast, a strong irrelevant motion is detected by the higher control-systems and therefore suppressed, in order to better focus on the task at hand.
The results show that the expected TIPL effect is more likely to occur in the later stages of training. Interestingly, this applies to either direction, that is, both learning with threshold exposure and suppression of learning with suprathreshold exposure show similar temporal dynamics. This differs from learning in VPL, which may occur fast (e.g. Fahle et al., 1995; Yotsumoto et al., 2008). Possibly, in VPL attentional and control mechanisms enhance the relevant stimuli, thereby stimulating and/or prioritizing its learning. Since no enhancing mechanisms are not involved with (sub)threshold task-irrelevant stimuli, it may take much longer for TIPL to occur. Accordingly, it is expected that the associated plasticity with TIPL would occur in the later stages of training as well. With regard to suprathreshold irrelevant stimuli, the brain may more effectively suppress the irrelevant coherent motion
with more training, which may coincide with a better performance on the task at hand. However, these speculations remain yet to be investigated.
The current pilot study showed a similar trend as previous studies (Chang et al., 2014; Tsushima et al., 2008). It can therefore be concluded that the TIPL effect in either direction (learning with threshold exposure vs. suppression with suprathreshold exposure) is a fairly robust phenomenon. However, there are some limitations of the current study. First, a part of the experiment was conducted remotely and in different environments, which may possibly have confounded the results. However, similar results were found in the different
experimental settings. The only observed difference was a lower threshold for motion
visibility threshold prior to training in the laboratory setting compared to the at-home setting, which can be explained by the finer laboratory setting. Importantly, finding the expected result in different experimental settings adds to the ecological validity of the effect. Second, and most important, the current pilot study is underpowered. Therefore, collecting additional data is warranted in order to detect a learning effect. Nevertheless, the results are promising enough to do future investigations with the current experimental design. The neurochemical mechanisms remain yet to be explored. This way, a more complete description of TIPL will be achieved to further understand what and when something is learned.
Literature
Ahissar, M., & Hochstein, S. (1993). Attentional control of early perceptual learning. Proceedings of the National Academy of Sciences, 90(12), 5718-5722.
Ahissar, M., & Hochstein, S. (1997). Task difficulty and the specificity of perceptual learning. Nature, 387(6631), 401.
Ahissar, M., & Hochstein, S. (2004). The reverse hierarchy theory of visual perceptual learning. Trends in cognitive sciences, 8(10), 457-464.
Ball, K., & Sekuler, R. (1987). Direction-specific improvement in motion discrimination. Vision research, 27(6), 953-965.
Bang, J. W., Shibata, K., Frank, S. M., Walsh, E. G., Greenlee, M. W., Watanabe, T., & Sasaki, Y. (2018). Consolidation and reconsolidation share behavioural and neurochemical mechanisms. Nature human behaviour, 2(7), 507-513.
Born, R. T., & Bradley, D. C. (2005). Structure and function of visual area MT. Annu. Rev. Neurosci., 28, 157- 189.
Brainard, D. H. (1997). The psychophysics toolbox. Spatial vision, 10(4), 433-436.
Chang, L. H., Shibata, K., Andersen, G. J., Sasaki, Y., & Watanabe, T. (2014). Age-related declines of stability in visual perceptual learning. Current Biology, 24(24), 2926-2929.
Choi, H., Seitz, A. R., & Watanabe, T. (2009). When attention interrupts learning: inhibitory effects of attention on TIPL. Vision research, 49(21), 2586-2590.
Chung, S. T., Legge, G. E., & Cheung, S. H. (2004). recognition and reading speed in peripheral vision benefit from perceptual learning. Vision research, 44(7), 695-709.
Cohen Kadosh, K., Krause, B., King, A. J., Near, J., & Cohen Kadosh, R. (2015). Linking GABA and glutamate levels to cognitive skill acquisition during development. Human brain mapping, 36(11), 4334-4345. Fahle, M., Edelman, S., & Poggio, T. (1995). Fast perceptual learning in hyperacuity. Vision research, 35(21),
3003-3013.
Faul, F., Erdfelder, E., Lang, A. G., & Buchner, A. (2007). G* Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior research methods, 39(2), 175-191.
Floyer-Lea, A., Wylezinska, M., Kincses, T., & Matthews, P. M. (2006). Rapid modulation of GABA
concentration in human sensorimotor cortex during motor learning. Journal of neurophysiology, 95(3), 1639-1644.
Frangou, P., Correia, M., & Kourtzi, Z. (2018). GABA, not BOLD, reveals dissociable learning-dependent plasticity mechanisms in the human brain. eLife, 7, e35854.
Frank, S. M., Bründl, S., Frank, U., Sasaki, Y., Greenlee, M.W., & Watanabe, T. (under revision) Fundamental differences in visual perceptual learning between children and adults.
Frank, S. M., Greenlee, M. W., & Tse, P. U. (2018). Long time no see: Enduring behavioral and neuronal changes in perceptual learning of motion trajectories 3 years after training. Cerebral Cortex, 28(4), 1260-1271.
invariance. Vision research, 40(5), 473-484.
Furmanski, C. S., Schluppeck, D., & Engel, S. A. (2004). Learning strengthens the response of primary visual cortex to simple patterns. Current Biology, 14(7), 573-578.
Gilbert, C. D., Sigman, M., & Crist, R. E. (2001). The neural basis of perceptual learning. Neuron, 31(5), 681- 697.
Grossberg, S. (1982). How does a brain build a cognitive code?. In Studies of mind and brain (pp. 1-52). Springer, Dordrecht.
Karni, A., & Sagi, D. (1991). Where practice makes perfect in texture discrimination: evidence for primary visual cortex plasticity. Proceedings of the National Academy of Sciences, 88(11), 4966-4970. Kim, D., Ling, S., & Watanabe, T. (2015). Dual mechanisms governing reward-driven perceptual
learning. F1000Research, 4.
Kim, S., Stephenson, M. C., Morris, P. G., & Jackson, S. R. (2014). tDCS-induced alterations in GABA concentration within primary motor cortex predict motor learning and motor memory: a 7 T magnetic resonance spectroscopy study. Neuroimage, 99, 237-243.
Kourtzi, Z., Betts, L. R., Sarkheil, P., & Welchman, A. E. (2005). Distributed neural plasticity for shape learning in the human visual cortex. PLoS biology, 3(7).
Law, C. T., & Gold, J. I. (2008). Neural correlates of perceptual learning in a sensory-motor, but not a sensory, cortical area. Nature neuroscience, 11(4), 505.
Lu, Z. L., Hua, T., Huang, C. B., Zhou, Y., & Dosher, B. A. (2011). Visual perceptual learning. Neurobiology of learning and memory, 95(2), 145-151.
Maertens, M., & Pollmann, S. (2005). fMRI reveals a common neural substrate of illusory and real contours in V1 after perceptual learning. Journal of cognitive neuroscience, 17(10), 1553-1564.
Mermillod, M., Bugaiska, A., & Bonin, P. (2013). The stability-plasticity dilemma: Investigating the continuum from catastrophic forgetting to age-limited learning effects. Frontiers in psychology, 4, 504.
Nikolova, S., Stark, S. M., & Stark, C. E. (2017). 3T hippocampal glutamate-glutamine complex reflects verbal memory decline in aging. Neurobiology of aging, 54, 103-111.
Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial vision, 10(4), 437-442.
Pilly, P. K., Grossberg, S., & Seitz, A. R. (2010). Low-level sensory plasticity during task-irrelevant perceptual learning: evidence from conventional and double training procedures. Vision research, 50(4), 424-432.
Recanzone, G. H., Schreiner, C. E., & Merzenich, M. M. (1993). Plasticity in the frequency representation of primary auditory cortex following discrimination training in adult owl monkeys. Journal of
Neuroscience, 13(1), 87-103.
Sasaki, Y., Nanez, J. E., & Watanabe, T. (2010). Advances in visual perceptual learning and plasticity. Nature Reviews Neuroscience, 11(1), 53-60.
Schwartz, S., Maquet, P., & Frith, C. (2002). Neural correlates of perceptual learning: a functional MRI study of visual texture discrimination. Proceedings of the National Academy of Sciences, 99(26), 17137-17142. Schoups, A., Vogels, R., Qian, N., & Orban, G. (2001). Practising orientation identification improves
orientation coding in V1 neurons. Nature, 412(6846), 549-553.
Seitz, A. R., Kim, D., & Watanabe, T. (2009). Rewards evoke learning of unconsciously processed visual stimuli in adult humans. Neuron, 61(5), 700-707.
Seitz, A. R., & Watanabe, T. (2003). Psychophysics: Is subliminal learning really passive?. Nature, 422(6927), 36.
Seitz, A., & Watanabe, T. (2005). A unified model for perceptual learning. Trends in cognitive sciences, 9(7), 329-334.
Seitz, A. R., & Watanabe, T. (2009). The phenomenon of task-irrelevant perceptual learning. Vision research, 49(21), 2604-2610.
Shibata, K., Chang, L. H., Kim, D., Náñez Sr, J. E., Kamitani, Y., Watanabe, T., & Sasaki, Y. (2012). Decoding reveals plasticity in V3A as a result of motion perceptual learning. PLoS one, 7(8).
Shibata, K., Sasaki, Y., Bang, J. W., Walsh, E. G., Machizawa, M. G., Tamaki, M., ... & Watanabe, T. (2017). Overlearning hyperstabilizes a skill by rapidly making neurochemical processing
inhibitory-dominant. Nature neuroscience, 20(3), 470-475.
Shibata, K., Watanabe, T., Sasaki, Y., & Kawato, M. (2011). Perceptual learning incepted by decoded fMRI neurofeedback without stimulus presentation. Science, 334(6061), 1413-1415.
Shiu, L. P., & Pashler, H. (1992). Improvement in line orientation discrimination is retinally local but dependent on cognitive set. Perception & psychophysics, 52(5), 582-588.
Sigman, M., Pan, H., Yang, Y., Stern, E., Silbersweig, D., & Gilbert, C. D. (2005). Top-down reorganization of activity in the visual pathway after learning a shape identification task. Neuron, 46(5), 823-835. Stagg, C. J., Bachtiar, V., & Johansen-Berg, H. (2011). The role of GABA in human motor learning. Current
Tootell, R. B., Reppas, J. B., Kwong, K. K., Malach, R., Born, R. T., Brady, T. J., ... & Belliveau, J. W. (1995). Functional analysis of human MT and related visual cortical areas using magnetic resonance
imaging. Journal of Neuroscience, 15(4), 3215-3230.
Tsushima, Y., Sasaki, Y., & Watanabe, T. (2006). Greater disruption due to failure of inhibitory control on an ambiguous distractor. Science, 314(5806), 1786-1788.
Tsushima, Y., Seitz, A. R., & Watanabe, T. (2008). Task-irrelevant learning occurs only when the irrelevant feature is weak. Current biology, 18(12), R516-R517.
Watanabe, T., Náñez Sr, J. E., Koyama, S., Mukai, I., Liederman, J., & Sasaki, Y. (2002). Greater plasticity in lower-level than higher-level visual motion processing in a passive perceptual learning task. Nature neuroscience, 5(10), 1003.
Watanabe, T., Náñez, J. E., & Sasaki, Y. (2001). Perceptual learning without perception. Nature, 413(6858), 844.
Watanabe, T., & Sasaki, Y. (2015). Perceptual learning: toward a comprehensive theory. Annual review of psychology, 66, 197-221.
Yang, T., & Maunsell, J. H. (2004). The effect of perceptual learning on neuronal responses in monkey visual area V4. Journal of Neuroscience, 24(7), 1617-1626.
Yotsumoto, Y., Chang, L. H., Ni, R., Pierce, R., Andersen, G. J., Watanabe, T., & Sasaki, Y. (2014). White matter in the older brain is more plastic than in the younger brain. Nature Communications, 5, 5504. Yotsumoto, Y., Watanabe, T., & Sasaki, Y. (2008). Different dynamics of performance and brain activation in
