In the very early training stages of many perceptual tasks, an observer is uncertain about the specifics (e.g., size, orientation) of the signal and occasionally an observer's performance will get worse over a number of trials before starting to improve. What causes this effect? We investigate this effect using a rapid perceptual learning paradigm (RPL; Abbey et al., VSS, 2001). In RPL paradigm used in this study, sessions consisted of sets of 4 learning trials. At the beginning of a learning set, 1 of 4 possible targets (elongated oriented Gaussians) were randomly chosen and used throughout the set. On each trial, the target appeared randomly at one of 8 locations in visual noise. The observer's task was to localize the target. At the end of the 4th learning trial of each set, the observer also had to identify the target. The RPL paradigm allows to compare the amount of learning in humans to that of an optimal Bayesian observer which updates the weightings assigned to the perceptual templates as the learning trials progress. We investigated how the learning is affected by the observer's correct or incorrect identification of the target at the end of the 4th learning trial. All three observers studied showed an improvement in localization performance with learning trial number for the learning sets with correct target identification (∼ 8.7 % from the 1st to the 4th learning trial) and also across all learning sets (∼ 7%). However, learning sets in which the observers incorrectly identified the target resulted in a decrease in localization performance (∼ 8 %). These results suggest that when an observer is uncertain about the signal, occasionally the external noise (and potentially internal noise) provides the observer with evidence for the presence of an erroneous target. This causes the observer to monitor the incorrect template (as evidenced by the incorrect target identification) and produces negative learning.

Introduction: Many studies have shown how feedback can improve perceptual learning (e.g. Herzog and Fahle, 1997). Less is known about how well humans use feedback. To investigate this question, we use an experimental paradigm (rapid perceptual learning, RPL; Abbey et al., 2001; Eckstein et al., 2002) in which an ideal observer learns from trial to trial. The framework allows us to compare the amount of learning with and without feedback in the human observer to the maximal possible learning assessed by the ideal observer. Methods: In the present RPL paradigm a learning set consisted of 8 trials. One out of the 26 letters from the English alphabet was randomly chosen and remained as a target throughout a learning set. On each trial, the target letter appeared randomly in 1 out of 8 locations embedded in image noise. The observers had to localize the target on each trial and identify it on the last trial of the learning set. There were two blocked conditions: 1) no feedback, 2) feedback about the target location provided with a post-cue following the observers' localization response (post-cue feedback). Observers participated in 1200 learning sets. Results: Human localization performance across the 8 learning trials increased significantly for both conditions (averaged across conditions and observers = 8 %). Learning was greater for the feedback condition than the no feedback condition consistent with previous studies. Overall human efficiency (i.e. the squared ratio of the ideal observer contrast threshold and the human contrast threshold) decreased (>>7 %) from the first to the last learning trial for both conditions suggesting that humans learn less than the ideal observer. Efficiency for the feedback condition reached its lowest point in the 2nd learning trial suggesting that the ideal observer learns faster with feedback than humans. Conclusion: Humans use of feedback in a perceptual learning task is imperfect and slow compared to the ideal observer.

Human cue validity effects in many cueing tasks can be explained with a weighted linear (Kinchla, et al., P & P, 1995) or Bayesian (Shimozaki, et al., 2001, ARVO; Eckstein, et al., JOV, 2001) integration model that preferentially weights the cued location, without proposing either an improved quality of processing at the cued location, or limited attentional resources. Precue effects with a 100% valid simultaneous cue (Dosher & Lu, VR, 2000), however, are not predicted by a Bayesian (ideal) observer, which ignores the precue and considers only the test location indicated by the simultaneous cue. One explanation of these precueing effects is that attention improves the tuning of the perceptual template (or excludes irrelevant external noise) at the test location when the precue and simultaneous cue agree. Another hypothesis is that observers (unlike an ideal observer) are unable to ignore information from the precued location when the precue is invalid (i.e., attentional leaking). Classification images were used to estimate human perceptual templates at the precued (invalid trials) and test locations (valid and invalid). Two observers (LL and JT) participated in a cued contrast discrimination of a 50 ms Gaussian signal in image noise at four possible locations, with a 150 ms precue (62.5% valid) and a simultaneous cue (100% valid). LL had large precue effects (d′ 0.5), and her invalid precued classification image was not significantly different from the test location classification image. JT had no precue effects, and his invalid precued classification image did not differ from zero. Also, for both observers, no difference was found for the test location classification images in the valid and invalid precued trials. For this task, we conclude that precueing validity effects (for LL) were due to the inability to ignore information at the invalid precue location (attentional leaking), and not the improved tuning of perceptual templates at the test location.

Purpose: In the context of a visual task, the notion of perceptual learning is often used to explain sequential improvements in task performance. In this work, we use the ideal observer as a benchmark for measuring the efficiency of perceptual learning in humans. In the task we investigate, the ideal observer “learns” by exploiting redundant information in earlier trials to optimally weight feature responses. The goal is to understand how stimulus presentation and feedback influence the efficiency of human perceptual learning. Methods: We conducted a series of forced-choice localization experiments using noisy images in which a target appeared randomly in one of eight locations. Each experiment consisted of sets of four “learning” trials. At the beginning of a set of learning trials, one of four possible signals was chosen at random and used throughout the learning set. We investigated the effect of different stimulus presentation times (unlimited vs. 200ms) and different feedback methods (location − 200ms, stimulus + location − 200ms, stimulus + location − unlimited) on sequential task performance. We compared human observer performance to the performance of the Bayesian ideal observer in this task. The ideal observer optimally updates its prior information about the signal profile throughout the learning trials. Results: In all cases, our observers showed improved performance going from the first learning trial to the fourth. This indicates that learning did indeed occur in our observers. However, efficiency with respect to the ideal observer dropped after the first learning trial, except in the condition with unlimited presentation time. Conclusions: In general, efficiency with respect to the ideal observer is relatively constant across learning trials. However, for short stimulus presentation times, efficiency decreases after the first trial, indicating a slower rate of learning.

Behavioral studies using reports of target items after a cue from a rapidly presented stream of potential targets (i.e., RSVP, Weichselgartner & Sperling, 1987) and studies measuring event related potentials (ERP, Hillyard et al, 1998) both indicate a time course of attentional gating of about 100 ms. In the present study, the temporal dynamics of attentional shifts to a peripheral cued location were examined in a contrast discrimination task. In particular we assess the earliest time from cue onset in which information can be used at the attended location. We estimate this use of information by the spatiotemporal analog of the classification image technique (classification movie, Xing and Ahumada, 2002). Methods: Two observers had to detect the presence of a Gaussian (stdev= 8.2 min) contrast increment (yes/no task; 50% probability of target presence) embedded in a temporal sequence of independent frames of spatially uncorrelated noise (frame rate= 40 f/s) displayed for 450 ms. The target was randomly located in one of eight locations equidistant from fixation at an eccentricity of 4.6 deg. A 100% valid simultaneous cue indicated the probable location of the target. The noise fields at the cued location on the signal absent trials (false alarms and correct rejections) were used to estimate the spatial perceptual filters of the observer at the cued location at different temporal intervals from cue onset (classification movies). Results: Observers showed significant estimated perceptual filters at the cued location within 25–50 ms. Conclusions: Our results suggest the use of information at an attended location begins 25 to 50 ms after a peripheral cue to shift attention, earlier than estimates from previous studies.

Purpose: Some studies of learning have found that easier tasks enhance the learning process (e.g. Ahissar & Hochstein , Nature 1997). We investigate the influence of signal contrast on very rapid learning effects using an experimental paradigm in which an ideal observer learns from trial to trial. The framework allows us to compare the amount of learning in human observers to maximal learning assessed by the ideal observer across stimulus contrast. Methods: We conducted 8-alternative forced-choice localization experiments over contrasts of 12% to 17%. Each experiment consisted of blocks of four trials in which one of four possible signals (oriented bright or dark Gaussians, embedded in noise) was chosen at random and used throughout the block. Within a block, signal location was randomized, but signal identity was constant, allowing observers (human and ideal) to use previous trials within the block to “learn” the signal profile and thus improve their strategy for localization. The ideal observer allows for the computation of statistical efficiency. In addition, learning efficiency can be determined by comparing contrast thresholds (matched to human observer performance) of the ideal learner with an observer that is ideal except that it does not use information from previous trials (and hence is unable to learn). Results: We observed performance improvements of 3% to 16% in percent correct over the four learning trials. In general, statistical efficiency was relatively constant across learning trials, but as the stimulus contrast increased, it increased by 30% to 50%. Over this same range of contrasts, learning efficiency was highest at low contrast and fell off as contrast increased. Conclusions: Higher signal contrasts - which we have used to manipulate task difficulty - lead to a general improvement in statistical efficiency. However, in terms of learning efficiency, our results indicate that observers learn relatively more in difficult tasks than easy ones.

We used classification images (A. J. Ahumada, Jr., & J. Lovell, 1971) to estimate the perceptual filter in a task designed to assess both local and nonlocal effects upon contrast detection/discrimination. Three observers performed a yes/no detection or discrimination task of a uniform circular decrement (radius = 0.68 deg) near threshold presented for 100 to 400 ms. Stimuli were presented in ring image noise that either covered the signal and an annular surrounding area (out to 1.36 deg), or only the surrounding annular area (out to 1.36 deg). Both the signal and the annular surround appeared on a uniform background. With ring noise over both the signal and surround, the amplitudes of the classification images in the signal area decreased as radial distance increased from the signal/surround border, and no effect of the surround was found. With ring noise only in the surround, classification images indicated noncontiguous effects at both the signal/surround border (local) and the surround/background border (nonlocal). The spatial extents of the nonlocal effects (< 0.07 deg) were smaller than local effects (0.25 deg), whereas the peak amplitudes of the local and nonlocal effects were comparable. These results suggest that the nonlocal effects were smaller than the local effects, and that the smaller effects would be due to smaller effective areas, as opposed to smaller amplitudes over the same area. Little or no change was found in the classification images across stimulus duration, suggesting that both the local and nonlocal processes found in this study were completed within 100 ms.

Hemineglect describes patients with unilateral brain damage that tend to ignore the contralesional visual field. While neglect most commonly occurs with parietal lesions, damage to other areas may also cause neglect (see Karnath, Milner, & Vallar, 2002), possibly by different mechanisms. To identify potentially different causes of neglect, the perceptual templates in a cueing task were estimated for three normal observers and two male right-hemisphere lesioned patients (with previous histories of hemineglect) by correlating observers' responses with the image noise leading to those responses (‘classification images’). Observers performed a yes/no contrast discrimination of a signal appearing at one of two locations (2.5 deg left and right from center). Prior to the stimulus (140 ms), a peripheral precue (140 ms) indicated the signal location with 80% validity. The signal was a 3×3 ‘white X’ checkerboard (1.5 deg), with Gaussian image noise added to each of the 9 checkerboard squares. As expected, the cueing effects and classification images for the normal observers showed no hemispheric differences. As predicted from other studies of hemineglect (e.g., Posner, et al., 1984), the patients (CM, age 85; HL, age 69) had larger cueing effects with right-sided cues (contralesional invalid), compared to left-sided cues (contralesional valid). For CM, the contralesional classification images were positively correlated with the signal with contralesional cues, and negatively correlated with the signal with ipsilesional cues, indicating a severely suboptimal attentional system. For HL, the contralesional classification images were uncorrelated with the signal, regardless of the cue side. HL did not appear to utilize contralesional information, consistent with either a residual visual (hemianopia) or attentional sensitivity loss. In conclusion, classification images successfully distinguished between two different mechanisms of neglect within these two patients.

Human performance in visual detection, discrimination, identification, and search tasks typically improves with practice. Psychophysical studies suggest that perceptual learning is mediated by an enhancement in the coding of the signal, and physiological studies suggest that it might be related to the plasticity in the weighting or selection of sensory units coding task relevant information (learning through attention optimization). We propose an experimental paradigm (optimal perceptual learning paradigm) to systematically study the dynamics of perceptual learning in humans by allowing comparisons to that of an optimal Bayesian algorithm and a number of suboptimal learning models. We measured improvement in human localization (eight-alternative forced-choice with feedback) performance of a target randomly sampled from four elongated Gaussian targets with different orientations and polarities and kept as a target for a block of four trials. The results suggest that the human perceptual learning can occur within a lapse of four trials (<1 min) but that human learning is slower and incomplete with respect to the optimal algorithm (23.3% reduction in human efficiency from the 1st-to-4th learning trials). The greatest improvement in human performance, occurring from the 1st-to-2nd learning trial, was also present in the optimal observer, and, thus reflects a property inherent to the visual task and not a property particular to the human perceptual learning mechanism. One notable source of human inefficiency is that, unlike the ideal observer, human learning relies more heavily on previous decisions than on the provided feedback, resulting in no human learning on trials following a previous incorrect localization decision. Finally, the proposed theory and paradigm provide a flexible framework for future studies to evaluate the optimality of human learning of other visual cues and/or sensory modalities.

Human performance during visual search typically improves when spatial cues indicate the possible target locations. In many instances, the performance improvement is quantitatively predicted by a Bayesian or quasi-Bayesian observer in which visual attention simply selects the information at the cued locations without changing the quality of processing or sensitivity and ignores the information at the uncued locations. Aside from the general good agreement between the effect of the cue on model and human performance, there has been little independent confirmation that humans are effectively selecting the relevant information. In this study, we used the classification image technique to assess the effectiveness of spatial cues in the attentional selection of relevant locations and suppression of irrelevant locations indicated by spatial cues. Observers searched for a bright target among dimmer distractors that might appear (with 50% probability) in one of eight locations in visual white noise. The possible target location was indicated using a 100% valid box cue or seven 100% invalid box cues in which the only potential target locations was uncued. For both conditions, we found statistically significant perceptual templates shaped as differences of Gaussians at the relevant locations with no perceptual templates at the irrelevant locations. We did not find statistical significant differences between the shapes of the inferred perceptual templates for the 100% valid and 100% invalid cues conditions. The results confirm the idea that during search visual attention allows the observer to effectively select relevant information and ignore irrelevant information. The results for the 100% invalid cues condition suggests that the selection process is not drawn automatically to the cue but can be under the observers' voluntary control.

The eye movements of two patients with parietal lobe lesions and four normal observers were measured while they performed a visual search task with naturalistic objects. Patients were slower to perform the task than the normal observers, and the patients had more fixations per trial, longer latencies for the first saccade during the visual search, and less accurate first and second saccades to the target locations during the visual search. The increases in response times for the patients compared to the normal observers were best predicted by increases in the number of fixations. In order to investigate the effects of spatial memory on search performance, in some trials observers saw a preview of the search display. The patients appeared to have difficulty using previously viewed information, unlike normal observers who benefit from the preview. This suggests a spatial memory deficit. The patients' deficits are consistent with the hypothesis that the parietal cortex has a role in the selection of targets for saccades, in memory for target location.

In a task in which the observer must detect a signal at two locations, presenting a precue that predicts the location of a signal leads to improved performance with a valid cue (signal location matches the cue), compared to an invalid cue (signal location does not match the cue). The cue validity effect has often been explained with a limited capacity attentional mechanism improving the perceptual quality at the cued location. Alternatively, the cueing effect can also be explained by unlimited capacity models that assume a weighted combination of noisy responses across the two locations. We compare two weighted integration models, a linear model and a sum of weighted likelihoods model based on a Bayesian observer. While qualitatively these models are similar, quantitatively they predict different cue validity effects as the signal-to-noise ratios (SNR) increase. To test these models, 3 observers performed in a cued discrimination task of Gaussian targets with an 80% valid precue across a broad range of SNR's. Analysis of a limited capacity attentional switching model was also included and rejected. The sum of weighted likelihoods model best described the psychophysical results, suggesting that human observers approximate a weighted combination of likelihoods, and not a weighted linear combination.

We describe a probit regression approach for maximum-likelihood (ML) estimation of a linear observer template from human-observer data in two-alternative forced-choice experiments. Like a previous approach to ML estimation in this problem [Abbey & Eckstein, Proc. SPIE, Vol. 4324, 2001], our approach does not make any assumptions about the distribution of the images. The previous approach utilized a regularizing prior distribution to control the degrees of freedom in the problem. In this work, we constrain the observer template to be represented by a limited number of linear features. Standard methods of probit regression are described for estimating the feature weights, and hence the observer templates.We have used this probit regression method to estimate human-observer templates for the detection of a small (5mm diameter) round simulated mass embedded in digitized mammograms. Our estimated templates for detecting the mass contain a band of heavily weighted spatial frequencies from 0.08 to 0.3 cycles/mm. We show comparisons between the human-observer template data, and the templates of a number of linear model observers that have been investigated as perceptual models of the human.