Signal but not noise changes with perceptual learning

Reduction of internal noise in auditory perceptual learning

This paper examines what mechanisms underlie auditory perceptual learning. Fifteen normal hearing adults performed two-alternative, forced choice, pure tone frequency discrimination for four sessions. External variability was introduced by adding a zero-mean Gaussian random variable to the frequency of each tone. Measures of internal noise, encoding efficiency, bias, and inattentiveness were derived using four methods (model fit, classification boundary, psychometric function, and double-pass consistency). The four methods gave convergent estimates of internal noise, which was found to decrease from 4.52 to 2.93 Hz with practice. No group-mean changes in encoding efficiency, bias, or inattentiveness were observed. It is concluded that learned improvements in frequency discrimination primarily reflect a reduction in internal noise. Data from highly experienced listeners and neural networks performing the same task are also reported. These results also indicated that auditory learning represents internal noise reduction, potentially through the re-weighting of frequency-specific channels.

Learning to identify crowded letters: does the learning depend on the frequency of training?

Performance for many visual tasks improves with training. The magnitude of improvement following training depends on the training task, number of trials per training session and the total amount of training. Does the magnitude of improvement also depend on the frequency of training sessions? In this study, we compared the learning effect for three groups of normally sighted observers who repeatedly practiced the task of identifying crowded letters in the periphery for six sessions (1000 trials per session), according to three different training schedules-one group received one session of training everyday, the second group received a training session once a week and the third group once every 2weeks. Following six sessions of training, all observers improved in their performance of identifying crowded letters in the periphery. Most importantly, the magnitudes of improvement were similar across the three training groups. The improvement was accompanied by a reduction in the spatial extent of crowding, an increase in the size of visual span and a reduction in letter-size threshold. The magnitudes of these accompanied improvements were also similar across the three training groups. Our finding that the effectiveness of visual perceptual learning is similar for daily, weekly and biweekly training has significant implication for adopting perceptual learning as an option to improve visual functions for clinical patients.

Neural correlates of face gender discrimination learning

Using combined psychophysics and event-related potentials (ERPs), we investigated the effect of perceptual learning on face gender discrimination and probe the neural correlates of the learning effect. Human subjects were trained to perform a gender discrimination task with male or female faces. Before and after training, they were tested with the trained faces and other faces with the same and opposite genders. ERPs responding to these faces were recorded. Psychophysical results showed that training significantly improved subjects' discrimination performance and the improvement was specific to the trained gender, as well as to the trained identities. The training effect indicates that learning occurs at two levels-the category level (gender) and the exemplar level (identity). ERP analyses showed that the gender and identity learning was associated with the N170 latency reduction at the left occipital-temporal area and the N170 amplitude reduction at the right occipital-temporal area, respectively. These findings provide evidence for the facilitation model and the sharpening model on neuronal plasticity from visual experience, suggesting a faster processing speed and a sparser representation of face induced by perceptual learning.

Learning to detect and combine the features of an object

To recognize an object, it is widely supposed that we first detect and then combine its features. Familiar objects are recognized effortlessly, but unfamiliar objects--like new faces or foreign-language letters--are hard to distinguish and must be learned through practice. Here, we describe a method that separates detection and combination and reveals how each improves as the observer learns. We dissociate the steps by two independent manipulations: For each step, we do or do not provide a bionic crutch that performs it optimally. Thus, the two steps may be performed solely by the human, solely by the crutches, or cooperatively, when the human takes one step and a crutch takes the other. The crutches reveal a double dissociation between detecting and combining. Relative to the two-step ideal, the human observer's overall efficiency for unconstrained identification equals the product of the efficiencies with which the human performs the steps separately. The two-step strategy is inefficient: Constraining the ideal to take two steps roughly halves its identification efficiency. In contrast, we find that humans constrained to take two steps perform just as well as when unconstrained, which suggests that they normally take two steps. Measuring threshold contrast (the faintness of a barely identifiable letter) as it improves with practice, we find that detection is inefficient and learned slowly. Combining is learned at a rate that is 4× higher and, after 1,000 trials, 7× more efficient. This difference explains much of the diversity of rates reported in perceptual learning studies, including effects of complexity and familiarity.

Asymmetric transfer of auditory perceptual learning

Perceptual skills can improve dramatically even with minimal practice. A major and practical benefit of learning, however, is in transferring the improvement on the trained task to untrained tasks or stimuli, yet the mechanisms underlying this process are still poorly understood. Reduction of internal noise has been proposed as a mechanism of perceptual learning, and while we have evidence that frequency discrimination (FD) learning is due to a reduction of internal noise, the source of that noise was not determined. In this study, we examined whether reducing the noise associated with neural phase locking to tones can explain the observed improvement in behavioral thresholds. We compared FD training between two tone durations (15 and 100 ms) that straddled the temporal integration window of auditory nerve fibers upon which computational modeling of phase locking noise was based. Training on short tones resulted in improved FD on probe tests of both the long and short tones. Training on long tones resulted in improvement only on the long tones. Simulations of FD learning, based on the computational model and on signal detection theory, were compared with the behavioral FD data. We found that improved fidelity of phase locking accurately predicted transfer of learning from short to long tones, but also predicted transfer from long to short tones. The observed lack of transfer from long to short tones suggests the involvement of a second mechanism. Training may have increased the temporal integration window which could not transfer because integration time for the short tone is limited by its duration. Current learning models assume complex relationships between neural populations that represent the trained stimuli. In contrast, we propose that training-induced enhancement of the signal-to-noise ratio offers a parsimonious explanation of learning and transfer that easily accounts for asymmetric transfer of learning.

Transfer of perceptual learning between different visual tasks

Practice in most sensory tasks substantially improves perceptual performance. A hallmark of this 'perceptual learning' is its specificity for the basic attributes of the trained stimulus and task. Recent studies have challenged the specificity of learned improvements, although transfer between substantially different tasks has yet to be demonstrated. Here, we measure the degree of transfer between three distinct perceptual tasks. Participants trained on an orientation discrimination, a curvature discrimination, or a 'global form' task, all using stimuli comprised of multiple oriented elements. Before and after training they were tested on all three and a contrast discrimination control task. A clear transfer of learning was observed, in a pattern predicted by the relative complexity of the stimuli in the training and test tasks. Our results suggest that sensory improvements derived from perceptual learning can transfer between very different visual tasks.

How Noisy is Lexical Decision?

Lexical decision is one of the most frequently used tasks in word recognition research. Theoretical conclusions are typically derived from a linear model on the reaction times (RTs) of correct word trials only (e.g., linear regression and ANOVA). Although these models estimate random measurement error for RTs, considering only correct trials implicitly assumes that word/non-word categorizations are without noise: words receive a yes-response because they have been recognized, and they receive a no-response when they are not known. Hence, when participants are presented with the same stimuli on two separate occasions, they are expected to give the same response. We demonstrate that this not true and that responses in a lexical decision task suffer from inconsistency in participants’ response choice, meaning that RTs of “correct” word responses include RTs of trials on which participants did not recognize the stimulus. We obtained estimates of this internal noise using established methods from sensory psychophysics (Burgess and Colborne, 1988). The results show similar noise values as in typical psychophysical signal detection experiments when sensitivity and response bias are taken into account (Neri, 2010). These estimates imply that, with an optimal choice model, only 83–91% of the response choices can be explained (i.e., can be used to derive theoretical conclusions). For word responses, word frequencies below 10 per million yield alarmingly low percentages of consistent responses (near 50%). The same analysis can be applied to RTs, yielding noise estimates about three times higher. Correspondingly, the estimated amount of consistent trial-level variance in RTs is only 8%. These figures are especially relevant given the recent popularity of trial-level lexical decision models using the linear mixed-effects approach (e.g., Baayen et al., 2008).

Image correlates of crowding in natural scenes

Visual crowding is the inability to identify visible features when they are surrounded by other structure in the peripheral field. Since natural environments are replete with structure and most of our visual field is peripheral, crowding represents the primary limit on vision in the real world. However, little is known about the characteristics of crowding under natural conditions. Here we examine where crowding occurs in natural images. Observers were required to identify which of four locations contained a patch of "dead leaves'' (synthetic, naturalistic contour structure) embedded into natural images. Threshold size for the dead leaves patch scaled with eccentricity in a manner consistent with crowding. Reverse correlation at multiple scales was used to determine local image statistics that correlated with task performance. Stepwise model selection revealed that local RMS contrast and edge density at the site of the dead leaves patch were of primary importance in predicting the occurrence of crowding once patch size and eccentricity had been considered. The absolute magnitudes of the regression weights for RMS contrast at different spatial scales varied in a manner consistent with receptive field sizes measured in striate cortex of primate brains. Our results are consistent with crowding models that are based on spatial averaging of features in the early stages of the visual system, and allow the prediction of where crowding is likely to occur in natural images.

The eye limits the brain's learning potential

The concept of a critical period for visual development early in life during which sensory experience is essential to normal neural development is now well established. However recent evidence suggests that a limited degree of plasticity remains after this period and well into adulthood. Here, we ask the question, "what limits the degree of plasticity in adulthood?" Although this limit has been assumed to be due to neural factors, we show that the optical quality of the retinal image ultimately limits the brain potential for change. We correct the high-order aberrations (HOAs) normally present in the eye's optics using adaptive optics, and reveal a greater degree of neuronal plasticity than previously appreciated.

The perception of a face is no more than the sum of its parts

When you see a person's face, how do you go about combining his or her facial features to make a decision about who that person is? Most current theories of face perception assert that the ability to recognize a human face is not simply the result of an independent analysis of individual features, but instead involves a holistic coding of the relationships among features. This coding is thought to enhance people's ability to recognize a face beyond what would be expected if each feature were shown in isolation. In the study reported here, we explicitly tested this idea by comparing human performance on facial-feature integration with that of an optimal Bayesian integrator. Contrary to the predictions of most current notions of face perception, our findings showed that human observers integrate facial features in a manner that is no better than would be predicted by their ability to use each individual feature when shown in isolation. That is, a face is perceived no better than the sum of its individual parts.

Learning acts on distinct processes for visual form perception in the human brain

Learning is known to facilitate our ability to detect targets in clutter and optimize brain processes for successful visual recognition. Previous brain-imaging studies have focused on identifying spatial patterns (i.e., brain areas) that change with learning, implicating occipitotemporal and frontoparietal areas. However, little is known about the interactions within this network that mediate learning-dependent improvement in complex perceptual tasks (i.e., discrimination of visual forms in clutter). Here we take advantage of the complementary high spatial and temporal resolution of simultaneous EEG-fMRI to identify the learning-dependent changes in spatiotemporal brain patterns that mediate enhanced behavioral sensitivity in the discrimination of global forms after training. We measured the observers' choices when discriminating between concentric and radial patterns presented in noise before and after training. Similarly, we measured the choices of a pattern classifier when predicting each stimulus from EEG-fMRI signals. By comparing the performance of human observers and classifiers, we demonstrated that learning alters sensitivity to visual forms and EEG-fMRI activation patterns related to distinct visual recognition processes. In particular, behavioral improvement after training was associated with changes in (1) early processes involved in the integration of global forms in higher occipitotemporal and parietal areas, and (2) later processes related to categorical judgments in frontal circuits. Thus, our findings provide evidence that learning acts on distinct visual recognition processes and shapes feedforward interactions across brain areas to support performance in complex perceptual tasks.

Evidence for increased internal noise in migraineurs for contrast and shape processing

Aim: Increased contrast-level dependent internal noise has been reported in migraine. This study aimed to investigate whether a general increase in internal noise impacted on other tasks thought to assess functioning in cortical area V1 and was evident in global contour coding (V4).

Methods: Eleven migraineurs (six with aura) and 12 headache-free controls completed three psychophysical tasks: (i) contrast detection, (ii) discrimination of the angle of a spiral path and (iii) detection of deformation from circularity. Internal noise estimates were obtained using an N-pass method that compared responses to repeated presentations of identical stimuli. Internal noise results in inconsistent responses across different runs.

Results: Migraineurs had significantly higher contrast thresholds when there was high external luminance noise. There were no other significant group differences in thresholds. Increased multiplicative noise associated with contrast processing was replicated and increased additive noise, which is independent of the visual input, was found for the global form task.

Conclusions: This study provides further evidence for increased multiplicative internal noise associated with contrast processing in migraineurs. However there is no generalised increase in internal noise in V1 as noise estimates for angular discrimination were normal. Increased additive internal noise was associated with the global shape task, co-occurring with increased efficiency.

Neural representations for object perception: structure, category, and adaptive coding

Object perception is one of the most remarkable capacities of the primate brain. Owing to the large and indeterminate dimensionality of object space, the neural basis of object perception has been difficult to study and remains controversial. Recent work has provided a more precise picture of how 2D and 3D object structure is encoded in intermediate and higher-level visual cortices. Yet, other studies suggest that higher-level visual cortex represents categorical identity rather than structure. Furthermore, object responses are surprisingly adaptive to changes in environmental statistics, implying that learning through evolution, development, and also shorter-term experience during adulthood may optimize the object code. Future progress in reconciling these findings will depend on more effective sampling of the object domain and direct comparison of these competing hypotheses.

Motion aftereffect duration is not changed by perceptual learning: evidence against the representation modification hypothesis

The representation modification hypothesis of perceptual learning attributes the practice-induced improvements in sensitivity and/or discriminability to changes in the early visual areas. We used motion aftereffects (MAE) to probe the representations of motion direction. In two experiments, four practice sessions on a fine direction-discrimination task caused large stimulus-specific improvements in d' but no significant stimulus-specific changes in either static or dynamic MAE duration at posttest relative to a pretest. Power analysis indicated that the data were approximately 100 times more likely given the hypothesis of no MAE change than the hypothesis of a 10% relative change. In light of converging evidence in the MAE literature, this suggests that little or no change occurred in the cortical representations of visual motion up to and including area MT. The task specificity of the learning effect challenges the representation modification hypothesis and supports an alternative-selective reweighting.

Increased internal noise cannot account for motion coherence processing deficits in migraine

Aim: This study aimed to revisit previous findings of superior processing of motion direction in migraineurs with a more stringent direction discrimination task and to investigate whether increased internal noise can account for motion processing deficits in migraineurs.

Methods: Groups of 13 migraineurs (4 with aura, 9 without aura) and 15 headache-free controls completed three psychophysical tasks: one detecting coherence in a motion stimulus, one discriminating the spiral angle in a glass pattern and another discriminating the spiral angle in a global-motion task. Internal noise estimates were obtained for all tasks using an N-pass method.

Results: Consistent with previous research, migraineurs had higher motion coherence thresholds than controls. However, there were no significant performance differences on the spiral global-motion and global-form tasks. There was no significant group difference in internal noise estimates associated with any of the tasks.

Conclusions: The results from this study suggest that variation in internal noise levels is not the mechanism driving motion coherence threshold differences in migraine. Rather, we argue that motion processing deficits may result from cortical changes leading to less efficient extraction of global-motion signals from noise.

Superior identification of familiar visual patterns a year after learning

Practice improves visual performance on simple tasks in which stimuli vary along one dimension. Such learning frequently is stimulus-specific and enduring, and has been associated with plasticity in striate cortex. It is unclear if similar lasting effects occur for naturalistic patterns that vary on multiple dimensions. We measured perceptual learning in identification tasks that used faces and textures, stimuli that engage multiple stages in visual processing. Performance improved significantly across 2 consecutive days of practice. More important, the effects of practice were remarkably stable across time: Improvements were maintained approximately 1 year later, and both the relative difficulty of identifying individual stimuli and individual differences in performance were essentially constant across sessions. Finally, the effects of practice were largely stimulus-specific. Our results suggest that the characteristics of perceptual learning are similar across a spectrum of stimulus complexities.

How inherently noisy is human sensory processing?

Like any physical information processing device, the human brain is inherently noisy: If a participant is presented with the same sensory stimulus multiple times and is asked to press one of two buttons in response to some property of the stimulus, the response may vary even though the stimulus did not. This response variability can be used to estimate the amount of so-called internal noise-that is, noise that is not present in the stimulus (such as random dynamic dots on the screen) but in the participant's brain. How large is this internally generated noise? We obtained >400 independent estimates on 40 participants for a range of protocols (yes/no, two-, four-, and eight-alternative forced choice), modalities (auditory and visual), attentional state, adaptation state, stimulus types (static, moving, stereoscopic), and other parameters (timing, size, contrast). Our final estimate at ~1.3 (units of external noise standard deviation) is generally somewhat larger than that previously inferred from smaller and less varied data sets. We discuss the impact of high levels of internal noise on a number of experimental and computational efforts aimed at understanding and characterizing human sensory processing.

Perceptual learning as improved probabilistic inference in early sensory areas

Extensive training on simple tasks such as fine orientation discrimination results in large improvements in performance, a form of learning known as perceptual learning. Previous models have argued that perceptual learning is due to either sharpening and amplification of tuning curves in early visual areas or to improved probabilistic inference in later visual areas (at the decision stage). However, early theories are inconsistent with the conclusions of psychophysical experiments manipulating external noise, whereas late theories cannot explain the changes in neural responses that have been reported in cortical areas V1 and V4. Here we show that we can capture both the neurophysiological and behavioral aspects of perceptual learning by altering only the feedforward connectivity in a recurrent network of spiking neurons so as to improve probabilistic inference in early visual areas. The resulting network shows modest changes in tuning curves, in line with neurophysiological reports, along with a marked reduction in the amplitude of pairwise noise correlations.

Subjective and objective learning effects dissociate in space and in time

Perceptual learning not only improves sensitivity, but it also changes our subjective experience. However, the question of how these two learning effects relate is largely unexplored. Here we investigate how subjects learn to see initially indiscriminable metacontrast-masked shapes. We find that sensitivity and subjective awareness increase with training. However, sensitivity and subjective awareness dissociate in space: Learning effects on performance are lost when the task is performed at an untrained location in another quadrant, whereas learning effects on subjective awareness are maintained. This finding indicates that improvements in shape sensitivity involve visual areas up to V4, whereas changes in subjective awareness involve other brain regions. Furthermore, subjective awareness dissociates from sensitivity in time: In an early phase of perceptual learning, subjects perform above chance on trials that they rate as subjectively invisible. Later, this phenomenon disappears. Subjective awareness is thus neither necessary nor sufficient for achieving above-chance objective performance.

Improving reading speed for people with central vision loss through perceptual learning

PURPOSE

Perceptual learning has been shown to be effective in improving visual functions in the normal adult visual system, as well as in adults with amblyopia. In this study, the feasibility of applying perceptual learning to enhance reading speed in people with long-standing central vision loss was evaluated.

METHODS

Six observers (mean age, 73.8) with long-standing central vision loss practiced an oral sentence-reading task, with words presented sequentially using rapid serial visual presentation (RSVP). A pre-test consisted of measurements of visual acuities, RSVP reading speeds for six print sizes, the location of the preferred retinal locus for fixation (fPRL), and fixation stability. Training consisted of six weekly sessions of RSVP reading, with 300 sentences presented per session. A post-test, identical with the pre-test, followed the training.

RESULTS

All observers showed improved RSVP reading speed after training. The improvement averaged 53% (range, 34-70%). Comparisons of pre- and post-test measurements revealed little changes in visual acuity, critical print size, location of the fPRL, and fixation stability.

CONCLUSIONS

The specificity of the learning effect, and the lack of changes to the fPRL location and fixation stability suggest that the improvements are not due to observers adopting a retinal location with better visual capability, or an improvement in fixation. Rather, the improvements are likely to represent genuine plasticity of the visual system despite the older ages of the observers, coupled with long-standing sensory deficits. Perceptual learning might be an effective way of enhancing visual performance for people with central vision loss.

Visual perceptual learning

Perceptual learning refers to the phenomenon that practice or training in perceptual tasks often substantially improves perceptual performance. Often exhibiting stimulus or task specificities, perceptual learning differs from learning in the cognitive or motor domains. Research on perceptual learning reveals important plasticity in adult perceptual systems, and as well as the limitations in the information processing of the human observer. In this article, we review the behavioral results, mechanisms, physiological basis, computational models, and applications of visual perceptual learning.

Learning to Discriminate Face Views

Although perceptual learning of simple visual features has been studied extensively and intensively for many years, we still know little about the mechanisms of perceptual learning of complex object recognition. In a series of seven experiments, human perceptual learning in discrimination of in-depth orientation of face view was studied using psychophysical methods. We trained subjects to discriminate face orientations around a face view (i.e., 30°) over eight daily sessions, which resulted in a significant improvement in sensitivity to the face view orientation. This improved sensitivity was highly specific to the trained orientation and persisted up to 6 mo. Different from perceptual learning of simple visual features, this orientation-specific learning effect could completely transfer across changes in face size, visual field, and face identity. A complete transfer also occurred between two partial face images that were mutually exclusive but constituted a complete face. However, the transfer of the learning effect between upright and inverted faces and between a face and a paperclip object was very weak. These results shed light on the mechanisms of the perceptual learning of face view discrimination. They suggest that the visual system had learned how to compute face orientation from face configural information more accurately and that a large amount of plastic changes took place at a level of higher visual processing where size-, location-, and identity-invariant face views are represented.

Spatial frequency discrimination learning in normal and developmentally impaired human vision

Perceptual learning effects demonstrate that the adult visual system retains neural plasticity. If perceptual learning holds any value as a treatment tool for amblyopia, trained improvements in performance must generalise. Here we investigate whether spatial frequency discrimination learning generalises within task to other spatial frequencies, and across task to contrast sensitivity. Before and after training, we measured contrast sensitivity and spatial frequency discrimination (at a range of reference frequencies 1, 2, 4, 8, 16 c/deg). During training, normal and amblyopic observers were divided into three groups. Each group trained on a spatial frequency discrimination task at one reference frequency (2, 4, or 8 c/deg). Normal and amblyopic observers who trained at lower frequencies showed a greater rate of within task learning (at their reference frequency) compared to those trained at higher frequencies. Compared to normals, amblyopic observers showed greater within task learning, at the trained reference frequency. Normal and amblyopic observers showed asymmetrical transfer of learning from high to low spatial frequencies. Both normal and amblyopic subjects showed transfer to contrast sensitivity. The direction of transfer for contrast sensitivity measurements was from the trained spatial frequency to higher frequencies, with the bandwidth and magnitude of transfer greater in the amblyopic observers compared to normals. The findings provide further support for the therapeutic efficacy of this approach and establish general principles that may help develop more effective protocols for the treatment of developmental visual deficits.

Ideal observer analysis of crowding and the reduction of crowding through learning

Crowding is a prominent phenomenon in peripheral vision where nearby objects impede one's ability to identify a target of interest. The precise mechanism of crowding is not known. We used ideal observer analysis and a noise-masking paradigm to identify the functional mechanism of crowding. We tested letter identification in the periphery with and without flanking letters and found that crowding increases equivalent input noise and decreases sampling efficiency. Crowding effectively causes the signal from the target to be noisier and at the same time reduces the visual system's ability to make use of a noisy signal. After practicing identification of flanked letters without noise in the periphery for 6 days, subjects' performance for identifying flanked letters improved (reduction of crowding). Across subjects, the improvement was attributable to either a decrease in crowding-induced equivalent input noise or an increase in sampling efficiency, but seldom both. This pattern of results is consistent with a simple model whereby learning reduces crowding by adjusting the spatial extent of a perceptual window used to gather relevant input features. Following learning, subjects with inappropriately large windows reduced their window sizes; while subjects with inappropriately small windows increased their window sizes. The improvement in equivalent input noise and sampling efficiency persists for at least 6 months.

Modeling mechanisms of perceptual learning with augmented Hebbian re-weighting

Using the external noise plus training paradigm, we have consistently found that two independent mechanisms, stimulus enhancement and external noise exclusion, support perceptual learning in a range of tasks. Here, we show that re-weighting of stable early sensory representations through Hebbian learning (Petrov et al., 2005, 2006) can generate performance patterns that parallel a large range of empirical data: (1) perceptual learning reduced contrast thresholds at all levels of external noise in peripheral orientation identification (Dosher & Lu, 1998, 1999), (2) training with low noise exemplars transferred to performance in high noise, while training with exemplars embedded in high external noise transferred little to performance in low noise (Dosher & Lu, 2005), and (3) pre-training in high external noise only reduced subsequent learning in high external noise, whereas pre-training in zero external noise left very little additional learning in all the external noise conditions (Lu et al., 2006). In the augmented Hebbian re-weighting model (AHRM), perceptual learning strengthens or maintains the connections between the most closely tuned visual channels and a learned categorization structure, while it prunes or reduces inputs from task-irrelevant channels. Reducing the weights on irrelevant channels reduces the contributions of external noise and additive internal noise. Manifestation of stimulus enhancement or external noise exclusion depends on the initial state of internal noise and connection weights in the beginning of a learning task. Both mechanisms reflect re-weighting of stable early sensory representations.

The neural basis of visual object learning

Object vision in human and nonhuman primates is often cited as a primary example of adult plasticity in neural information processing. It has been hypothesized that visual experience leads to single neurons in the monkey brain with strong selectivity for complex objects, and to regions in the human brain with a preference for particular categories of highly familiar objects. This view suggests that adult visual experience causes dramatic local changes in the response properties of high-level visual cortex. Here, we review the current neurophysiological and neuroimaging evidence and find that the available data support a different conclusion: adult visual experience introduces moderate, relatively distributed effects that modulate a pre-existing, rich and flexible set of neural object representations.

Pattern recognition in correlated and uncorrelated noise

This study examined how correlated, or filtered, noise affected efficiency for recognizing two types of signal patterns, Gabor patches and three-dimensional objects. In general, compared with the ideal observer, human observers were most efficient at performing tasks in low-pass noise, followed by white noise; they were least efficient in high-pass noise. Simulations demonstrated that contrast-dependent internal noise was likely to have limited human performance in the high-pass conditions for both signal types. Classification images showed that observers were likely adopting different strategies in the presence of low-pass versus white noise. However, efficiencies were underpredicted by the linear classification images and asymmetries were present in the classification subimages, indicating the influence of nonlinear processes. Response consistency analyses indicated that lower contrast-dependent internal noise contributed somewhat to higher efficiencies in low-pass noise for Gabor patches but not objects. Taken together, the results of these experiments suggest a complex interaction among signals, external noise spectra, and internal noise in determining efficiency in correlated and uncorrelated noise.

Learning motion: human vs. optimal Bayesian learner

We used the optimal perceptual learning paradigm (Eckstein, Abbey, Pham, & Shimozaki, 2004) to investigate the dynamics of human rapid learning processes in motion discrimination tasks and compare it to an optimal Bayesian learner. This paradigm consists of blocks of few trials defined by a set of target attributes, and it has been shown its ability to detect learning effects appearing as soon as after the first trial. In the present task a sequence consisting of four patches containing random-dot patterns is presented at four separate locations equidistant from a fixation point. On each trial, the random dots in three patches moved with a mean speed and the fourth, target patch, could move either with slower or faster mean speed. Observers' task was to indicate what speed, faster or slower, was present in the display. The mean direction of the target patch was kept invariant along a block of trials. Observers learned the target relevant motion direction through indirect feedback, leading to an improvement in speed identification performance ranging from 15% to 30% which is greater than previously studied contrast defined targets and faces. However, comparison to an ideal learner revealed incomplete or partial learning for the motion task which was lower than previously measured for contrast defined targets and faces. A sub-optimal model that included inefficiencies in the updating of motion direction weights due to memory effects could account for the human learning. Finally, the similarity of the rapid learning effect observed here for motion perception with that found for contrast defined targets for localization and identification tasks could be suggesting a general strategy for learning in the human visual system and some common limitations such as memory.

Visual learning for perceptual and categorical decisions in the human brain

Successful actions and interactions in the complex environments we inhabit entail making fast and optimal perceptual decisions. Extracting the key features from our sensory experiences and deciding how to interpret them is a computationally challenging task that is far from understood. Accumulating evidence suggests that the brain may solve this challenge by combining sensory information and previous knowledge about the environment acquired through evolution, development, and everyday experience. Here, we review the role of visual learning and experience-dependent plasticity in shaping decisions. We propose that learning plays an important role in translating sensory experiences to decisions and actions by shaping neural representations across cortical circuits in a task-dependent manner.

Perceptual learning as a potential treatment for amblyopia: a mini-review

Amblyopia is a developmental abnormality that results from physiological alterations in the visual cortex and impairs form vision. It is a consequence of abnormal binocular visual experience during the "sensitive period" early in life. While amblyopia can often be reversed when treated early, conventional treatment is generally not undertaken in older children and adults. A number of studies over the last twelve years or so suggest that Perceptual Learning (PL) may provide an important new method for treating amblyopia. The aim of this mini-review is to provide a critical review and "meta-analysis" of perceptual learning in adults and children with amblyopia, with a view to extracting principles that might make PL more effective and efficient. Specifically we evaluate: 1). What factors influence the outcome of perceptual learning? 2). Specificity and generalization - two sides of the coin. 3). Do the improvements last? 4). How does PL improve visual function? 5). Should PL be part of the treatment armamentarium? A review of the extant studies makes it clear that practicing a visual task results in a long-lasting improvement in performance in an amblyopic eye. The improvement is generally strongest for the trained eye, task, stimulus and orientation, but appears to have a broader spatial frequency bandwidth than in normal vision. Importantly, practicing on a variety of different tasks and stimuli seems to transfer to improved visual acuity. Perceptual learning operates via a reduction of internal neural noise and/or through more efficient use of the stimulus information by retuning the weighting of the information. The success of PL raises the question of whether it should become a standard part of the armamentarium for the clinical treatment of amblyopia, and suggests several important principles for effective perceptual learning in amblyopia.

How much practice is needed to produce perceptual learning?

We examined the amount of practice needed to improve performance on 10-AFC face- and texture identification tasks. On Day 1, subjects were grouped by amount of practice: a control group had 0 trials of practice, and several experimental groups had practice that ranged from 1 to 40 trials per condition. On Day 2, all groups performed 40 trials per condition of the trained task. The effect of practice was estimated by comparing performance across groups on Day 2. In both tasks, increasing practice was associated with greater learning, but surprisingly small amounts of practice were required to improve performance. In the face identification task, for example, only one trial per condition on Day 1 was required to increase performance relative to the control group at the start of testing on Day 2. In the texture identification task, five trials per condition on Day 1 were required to increase performance relative to the control group. In both tasks, the advantage associated with small amounts of practice declined during the Day 2 session due to larger within-session learning in the control group. Sleep had little to no effect on learning; performance depended primarily on the amount of preceding practice.

Shared mechanisms of perceptual learning and decision making

Perceptual decisions require the brain to weigh noisy evidence from sensory neurons to form categorical judgments that guide behavior. Here we review behavioral and neurophysiological findings suggesting that at least some forms of perceptual learning do not appear to affect the response properties of neurons that represent the sensory evidence. Instead, improved perceptual performance results from changes in how the sensory evidence is selected and weighed to form the decision. We discuss the implications of this idea for possible sites and mechanisms of training-induced improvements in perceptual processing in the brain.

Perceptual learning and representational learning in humans and animals

Traditionally, perceptual learning in humans and classical conditioning in animals have been considered as two very different research areas, with separate problems, paradigms, and explanations. However, a number of themes common to these fields of research emerge when they are approached from the more general concept of representational learning. To demonstrate this, I present results of several learning experiments with human adults and infants, exploring how internal representations of complex unknown visual patterns might emerge in the brain. I provide evidence that this learning cannot be captured fully by any simple pairwise associative learning scheme, but rather by a probabilistic inference process called Bayesian model averaging, in which the brain is assumed to formulate the most likely chunking/grouping of its previous experience into independent representational units. Such a generative model attempts to represent the entire world of stimuli with optimal ability to generalize to likely scenes in the future. I review the evidence showing that a similar philosophy and generative scheme of representation has successfully described a wide range of experimental data in the domain of classical conditioning in animals. These convergent findings suggest that statistical theories of representational learning might help to link human perceptual learning and animal classical conditioning results into a coherent framework.

Flexible learning of natural statistics in the human brain

The ability to detect and identify targets in cluttered scenes is a critical skill for survival and interactions. To solve this challenge the brain has optimized mechanisms for capitalizing on frequently occurring regularities in the environment. Although evolution and development have been suggested to shape the brain's architecture in a manner that resembles these natural statistics, we provide novel evidence that short-term experience in adulthood may modify the brain's functional organization to support integration of signals atypical of shape contours in natural scenes. Although collinearity is a prevalent principle for perceptual integration in natural scenes, we show that observers learn to exploit other image regularities (i.e., orthogonal alignments of segments at an angle to the contour path) that typically signify discontinuities. Combining behavioral and functional MRI measurements, we demonstrate that this flexible learning is mediated by changes in the neural representations of behaviorally relevant image regularities primarily in dorsal visual areas. These changes in neural sensitivity are in line with changes in perceptual sensitivity for the detection of orthogonal contours and are evident only in observers that show significant performance improvement. In contrast, changes in the activation extent in frontoparietal regions are evident independent of performance changes, may support the detection of salient regions, and modulate perceptual integration in occipitotemporal areas in a top-down manner. Thus experience at shorter timescales in adulthood supports the adaptive functional optimization of visual circuits for flexible interpretation of natural scenes.

Subordinate Categorization Enhances the Neural Selectivity in Human Object-selective Cortex for Fine Shape Differences

There is substantial evidence that object representations in adults are dynamically updated by learning. However, it is not clear to what extent these effects are induced by active processing of visual objects in a particular task context on top of the effects of mere exposure to the same objects. Here we show that the task does matter. We performed an event-related fMRI adaptation study in which we derived neural selectivity from a release of adaptation. We had two training conditions: “categorized objects” were categorized at a subordinate level based on fine shape differences (Which type of fish is this?), whereas “control objects” were seen equally often in a task context requiring no subordinate categorization (Is this a vase or not?). After training, the object-selective cortex was more selective for differences among categorized objects than for differences among control objects. This result indicates that the task context during training modulates the extent to which object selectivity is enhanced as a result of training.

Hebbian Reweighting on Stable Representations in Perceptual Learning

Perceptual learning is the improvement in perceptual task performance with practice or training. The observation of specificity in perceptual learning has been widely associated with plasticity in early visual cortex representations. Here, we review the evidence supporting the plastic reweighting of readout from stable sensory representations, originally proposed by Dosher & Lu (1998), as an alternative explanation of perceptual learning. A task-analysis that identifies circumstances in which specificity supports representation enhancement and those in which it implies reweighting provides a framework for evaluating the literature; reweighting is broadly consistent with the behavioral results and almost all of the physiological reports. We also consider the evidence that the primary mode of perceptual learning is through augmented Hebbian learning of the reweighted associations, which has implications for the role and importance of feedback. Feedback is not necessary for perceptual learning, but can improve it in some circumstances, and in some cases block feedback is also helpful - all effects that are generally compatible with an augmented Hebbian model (Petrov, Dosher, & Lu, 2005). The two principles of perceptual learning through reweighting evidence from stable sensory representations and of augmented Hebbian learning provide a theoretical structure for the consideration of issues such as task difficulty, task roving, and cuing in perceptual learning.

MECHANISMS OF PERCEPTUAL LEARNING

What is learned in perceptual learning? How does perceptual learning change the perceptual system? We investigate these questions using a systems analysis of the perceptual system during the course of perceptual learning using psychophysical methods and models of the observer. Effects of perceptual learning on an observer's performance are characterized by external noise tests within the framework of noisy observer models. We find evidence that two independent mechanisms, external noise exclusion and stimulus enhancement support perceptual learning across a range of tasks. We suggest that both mechanisms may reflect re-weighting of stable early sensory representations.

Sensory noise explains auditory frequency discrimination learning induced by training with identical stimuli

Thresholds in various visual and auditory perception tasks have been found to improve markedly with practice at intermediate levels of task difficulty. Recently, however, there have been reports that training with identical stimuli, which, by definition, were impossible to discriminate correctly beyond chance, could induce as much discrimination learning as could training with different stimuli. These surprising findings have been interpreted as evidence that discrimination learning can occur in the absence of perceived differences between stimuli and need not involve the fine-tuning of a discrimination mechanism. Here, we show that these counterintuitive findings of discrimination learning without discrimination can be understood simply by considering the effect of internal noise on sensory representations. Because of such noise, physically identical stimuli are unlikely to be perceived as being strictly identical. We show that, given empirically derived levels of sensory noise, perceived differences evoked by identical stimuli are actually not much smaller than those induced by the physical differences typically used in discrimination-learning experiments. We suggest that findings of discrimination learning with identical stimuli can be explained without implicating any fundamentally new learning mechanism.

Prolonged perceptual learning of positional acuity in adult amblyopia: perceptual template retuning dynamics

Amblyopia is a developmental abnormality that results in physiological alterations in the visual cortex and impairs form vision. It is often successfully treated by patching the sound eye in infants and young children, but is generally considered to be untreatable in adults. However, a number of recent studies suggest that repetitive practice of a visual task using the amblyopic eye results in improved performance in both children and adults with amblyopia. These perceptual learning studies have used relatively brief periods of practice; however, clinical studies have shown that the time-constant for successful patching is long. The time-constant for perceptual learning in amblyopia is still unknown. Here we show that the time-constant for perceptual learning depends on the degree of amblyopia. Severe amblyopia requires >50 h (approximately equal to 35,000 trials) to reach plateau, yielding as much as a five-fold improvement in performance at a rate of approximately equal to 1.5%/h. There is significant transfer of learning from the amblyopic to the dominant eye, suggesting that the learning reflects alterations in higher decision stages of processing. Using a reverse correlation technique, we document, for the first time, a dynamic retuning of the amblyopic perceptual decision template and a substantial reduction in internal spatial distortion. These results show that the mature amblyopic brain is surprisingly malleable, and point to more intensive treatment methods for amblyopia.

Mapping dynamic memories of gradually changing objects

Our brain is able to maintain a continuously updated memory representation of objects despite changes in their appearance over time (aging faces or objects, growing trees, etc.). Although this ability is crucial for cognition and behavior, it was barely explored. Here, we investigate this memory characteristic using a protocol emulating face transformation. Observers were presented with a sequence of faces that gradually transformed over many days, from a known face (source) to a new face (target), in presentations separated by other stimuli. This practice resulted in a drastic change in the memory and recognition of the faces. Although identification of the source and older face instances was reduced, recent face instances were increasingly identified as the source and rated as highly similar to the memory of the source. Using an object perturbation method, we estimated the corresponding memory shift, showing that memory patterns shifted from the source neighborhood toward the target. Our findings suggest that memory is updated to account for object changes over time while still keeping associations with past appearances. These experimental results are broadly compatible with a recently developed model of associative memory that assumes attractor dynamics with a learning rule facilitated by novelty, shown to hold when objects change gradually over short timescales.