Crowding reveals fundamental differences in local vs. global processing in humans and machines.

Feedforward Convolutional Neural Networks (ffCNNs) have become state-of-the-art models both in computer vision and neuroscience. However, human-like performance of ffCNNs does not necessarily imply human-like computations. Previous studies have suggested that current ffCNNs do not make use of global shape information. However, it is currently unclear whether this reflects fundamental differences between ffCNN and human processing or is merely an artefact of how ffCNNs are trained. Here, we use visual crowding as a well-controlled, specific probe to test global shape computations. Our results provide evidence that ffCNNs cannot produce human-like global shape computations for principled architectural reasons. We lay out approaches that may address shortcomings of ffCNNs to provide better models of the human visual system.

Investigating prospective memory via eye tracking: No evidence for a monitoring deficit in older adults.

Prospective memory (PM) refers to remembering to perform intended actions in the future. Older adults in particular have been shown to be negatively affected by PM tasks that require a high amount of attentional resources (i.e., nonfocal tasks). This age-related PM deficit has been attributed to reduced target monitoring in this age group older adults. However, this conclusion was based on indirect measures of monitoring, such as costs of the ongoing task. The present study set out to 1) investigate older adults' PM target monitoring by, for the first time, employing a direct measure (i.e., eye tracking), 2) assess differences in monitoring between PM tasks that differ in their focality, and 3) examine whether differences in PM monitoring can indeed explain older adults' reduced PM performance in nonfocal tasks. Results demonstrate that while older, but not younger adults, showed reduced performance in a nonfocal PM task, overt monitoring (eye movements) of these groups did not differ between focality conditions. Further analyses showed that older adults' performance was still reduced on the strategically more demanding task after controlling for overt target monitoring (i.e., including only trials in which the participant monitored). In contrast to indirect measures of cue monitoring, our findings illustrate that older adults' deficits on nonfocal PM tasks cannot (exclusively) be explained by reduced monitoring. Instead, processing that takes place after target monitoring are discussed as possible mechanisms underlying older adults' reduced PM performance in nonfocal tasks.

Different colors of light lead to different adaptation and activation as determined by high-density EEG.

Light adaptation is crucial for coping with the varying levels of ambient light. Using high-density electroencephalography (EEG), we investigated how adaptation to light of different colors affects brain responsiveness. In a within-subject design, sixteen young participants were adapted first to dim white light and then to blue, green, red, or white bright light (one color per session in a randomized order). Immediately after both dim and bright light adaptation, we presented brief light pulses and recorded event-related potentials (ERPs). We analyzed ERP response strengths and brain topographies and determined the underlying sources using electrical source imaging. Between 150 and 261 ms after stimulus onset, the global field power (GFP) was higher after dim than bright light adaptation. This effect was most pronounced with red light and localized in the frontal lobe, the fusiform gyrus, the occipital lobe and the cerebellum. After bright light adaptation, within the first 100 ms after light onset, stronger responses were found than after dim light adaptation for all colors except for red light. Differences between conditions were localized in the frontal lobe, the cingulate gyrus, and the cerebellum. These results indicate that very short-term EEG brain responses are influenced by prior light adaptation and the spectral quality of the light stimulus. We show that the early EEG responses are differently affected by adaptation to different colors of light which may contribute to known differences in performance and reaction times in cognitive tests.

The effect of spacing regularity on visual crowding.

Crowding limits peripheral visual discrimination and recognition: a target easily identified in isolation becomes impossible to recognize when surrounded by other stimuli, often called flankers. Most accounts of crowding predict less crowding when the target-flanker distance increases. On the other hand, the importance of perceptual organization and target-flanker coherence in crowding has recently received more attention. We investigated the effect of target-flanker spacing on crowding in multi-element stimulus arrays. We show that increasing the average distance between the target and the flankers does not always decrease the amount of crowding but can even sometimes increase it. We suggest that the regularity of inter-element spacing plays an important role in determining the strength of crowding: regular spacing leads to the perception of a single, coherent, texture-like stimulus, making judgments about the individual elements difficult.

Carbon outclasses wood in racket paddles: Ratings of expert and intermediate tennis players.

Wooden racket paddles were modified with rubber and carbon fibre laminates and their differences tested in terms of flexural, damping, and coefficient of restitution properties. Four rackets types were designed: a wood reference, wood with rubber, carbon fibre 0°, and carbon fibre 90°. Seven expert and eight intermediate tennis players tested the rackets. To determine which of the four rackets suited the players best, we asked the players to compare the rackets two by two. After each pair tested, participants had to fill out a 4-item questionnaire in which different aspects of the rackets' performance were judged. The most preferred racket was the 0° carbon fibre racket, followed by the 90° carbon fibre racket, the wood racket and, finally, the 1-mm rubber racket. Thus, rackets with the highest stiffness, least damping, and highest coefficient of restitution were the most preferred. Interestingly, although experts and intermediate players overall judged the rackets in very similar ways according to force, vibration, and control, they were sensitive to quite different physical characteristics of the rackets.

Gestalt factors modulate basic spatial vision.

Human perception of a stimulus varies depending on the context in which the stimulus is presented. Such contextual modulation has often been explained by two basic neural mechanisms: lateral inhibition and spatial pooling. In the present study, we presented observers with a vernier stimulus flanked by single lines; observers' ability to discriminate the offset direction of the vernier stimulus deteriorated in accordance with both explanations. However, when the flanking lines were part of a geometric shape (i.e., a good Gestalt), this deterioration strongly diminished. These findings cannot be explained by lateral inhibition or spatial pooling. It seems that Gestalt factors play an important role in contextual modulation. We propose that contextual modulation can be used as a quantitative measure to investigate the rules governing the grouping of elements into meaningful wholes.

Orientation specificity of learning vernier discriminations.

Orientation selective neurons in the primary visual cortex typically respond to a range of orientations that covers 20 degrees or more, while in psychophysical experiments, orientation bandwidth is often clearly narrower. Here, we measure the orientation specificity of perceptual learning for vernier discriminations. More than 70 observers, in separate groups, practiced a vernier discrimination task with a constant stimulus orientation. After a 1h session of training, the vernier was rotated by 2 degrees, 4 degrees, 10 degrees, 20 degrees, 45 degrees or 90 degrees. Improvement through training in the first session transferred to the second session (tested on the next day) up to 10 degrees of stimulus rotation. We found no transfer for rotations of 20 degrees, 45 degrees and 90 degrees. Hence, the orientation half-bandwidth of perceptual learning is around 15 degrees, leading to a bandwidth of 30 degrees and corresponding to that of single neurons in early visual cortices, while being narrower than that in higher cortical areas.

Spatial aspects of object formation revealed by a new illusion, shine-through.

When a vernier stimulus is presented for a short time and followed by a grating comprising five straight lines, the vernier remains invisible but may bequeath its offset to the grating (feature inheritance). For more than seven grating elements, the vernier is rendered visible as a shine-through element. However, shine-through depends strongly on the spatio-temporal layout of the grating. Here, we show that spatially inhomogeneous gratings diminish shine-through and vernier discrimination. Even subtle deviations, in the range of a few minutes of arc, matter. However, longer presentation times of the vernier regenerate shine-through. Feature inheritance and shine-through may become a useful tool in investigating such different topics as time course of information processing, feature binding, attention, and masking.

Shine-through: temporal aspects.

If a vernier stimulus precedes a grating for a very short time, the vernier either remains invisible, but may bequeath some of its properties to the grating (feature inheritance), or might shine through keeping its features - depending on the number of grating elements [Herzog, M. H. & Koch, C., 2001. Seeing properties of an invisible element: feature inheritance and shine-through. Proceedings of the National Academy of Science USA 98, 4271-4275]. Feature inheritance and shine-through represent two different states of feature binding [Herzog, M. H., Koch, C., & Fahle, M., Switching binding states. Visual Cognition (in press)], whereas shine-through depends in subtle ways on the spatial layout of the grating [Herzog, M. H., Fahle, M., & Koch, C., (2001). Spatial aspects of object formation revealed by a new illusion, shine-through Vision Research]. Here, we show that also temporal parameters of the grating influence shine-through. For example, a delayed presentation of certain grating elements can deteriorate performance dramatically.

Seeing properties of an invisible object: feature inheritance and shine-through.

We characterize a class of spatio-temporal illusions with two complementary properties. Firstly, if a vernier stimulus is flashed for a short time on a monitor and is followed immediately by a grating, the latter can express features of the vernier, such as its offset, its orientation, or its motion (feature inheritance). Yet the vernier stimulus itself remains perceptually invisible. Secondly, the vernier can be rendered visible by presenting gratings with a larger number of elements (shine-through). Under these conditions, subjects perceive two independent "objects" each carrying their own features. Transition between these two domains can be effected by subtle changes in the spatio-temporal layout of the grating. This should allow psychophysicists and electrophysiologists to investigate feature binding in a precise and quantitative manner.

Effects of biased feedback on learning and deciding in a vernier discrimination task.

We investigate the influence of biased feedback on decision and learning processes in a vernier discrimination task. Subjects adjust their decision criteria and hence their responses according to biased external feedback. However, they do not use learning processes to encode incorrectly classified stimuli. As soon as correct feedback is restored observers regain their original performance indicating an involvement of internal criteria. If the external feedback is switched off instead of being corrected, the rebound is less vigorous. The findings contradict predictions of supervised neural network models.

Modeling perceptual learning: difficulties and how they can be overcome.

We investigated the roles of feedback and attention in training a vernier discrimination task as an example of perceptual learning. Human learning even of simple stimuli, such as verniers, relies on more complex mechanisms than previously expected--ruling out simple neural network models. These findings are not just an empirical oddity but are evidence that present models fail to reflect some important characteristics of the learning process. We will list some of the problems of neural networks and develop a new model that solves them by incorporating top-down mechanisms. Contrary to neural networks, in our model learning is not driven by the set of stimuli only. Internal estimations of performance and knowledge about the task are also incorporated. Our model implies that under certain conditions the detectability of only some of the stimuli is enhanced while the overall improvement of performance is attributed to a change of decision criteria. An experiment confirms this prediction.

The role of feedback in learning a vernier discrimination task.

We compare improvement through training in vernier acuity under different feedback conditions in order to clarify the role of feedback during learning of a perceptual task and to test different (neural network) models of perceptual learning. Improvement of performance is measured in 49 observers under feedback, no feedback, uncorrelated feedback, partial feedback, and block feedback conditions. Correct feedback conditions yield a larger improvement of performance than manipulated and no feedback conditions. Providing feedback that is uncorrelated to the observers' responses prevents learning, while the effect of block feedback does not differ significantly from complete feedback. Our results cannot be explained by learning rules that depend exclusively on an external teacher or by models that propose learning in an exposure-dependent way with unsupervised learning rules but without top-down influences.