Learning features in a complex and changing environment: A distribution-based framework for visual attention and vision in general

What is the basis of ensemble subset selection?

The visual system can rapidly calculate the ensemble statistics of a set of objects; for example, people can easily estimate an average size of apples on a tree. To accomplish this, it is not always useful to summarize all the visual information. If there are various types of objects, the visual system should select a relevant subset: only apples, not leaves and branches. Here, we ask what kind of visual information makes a "good" ensemble that can be selectively attended to provide an accurate summary estimate. We tested three candidate representations: basic features, preattentive object files, and full-fledged bound objects. In four experiments, we presented a target and several distractors' sets of differently colored objects. We found that conditions where a target ensemble had at least one unique color (basic feature) provided ensemble averaging performance comparable to the baseline displays without distractors. When the target subset was defined as a conjunction of two colors or color-shape partly shared with distractors (so that they could be differentiated only as preattentive object files), subset averaging was also possible but less accurate than in the baseline and feature conditions. Finally, performance was very poor when the target subset was defined by an exact feature relationship, such as in the spatial conjunction of two colors (spatially bound object). Overall, these results suggest that distinguishable features and, to a lesser degree, preattentive object files can serve as the representational basis of ensemble selection, while bound objects cannot.

The influence of the tested item on serial dependence in perceptual decisions

Serial dependence in vision reflects how perceptual decisions can be biased by what we have recently perceived. Serial dependence studies test single items' effects on perceptual decisions. However, our visual world contains multiple objects at any given moment, so it's important to understand how past experiences affect not only a single object but also perception in a more general sense. Here we asked the question: What effect does a single item have when there is more than one subsequently presented test item? We displayed a single line (inducer) at the screen center, then either a single test-line or two simultaneous test-lines, varying in orientation space to the inducer. Next, participants reported test-line orientation using a left or right located response circle (to indicate which test-line should be reported). The results demonstrated that the inducer influenced subsequent perceptual judgments of a test-line even when two test-lines were presented. Distant items caused repulsive serial dependence, while close items caused attractive serial dependence. This shows how a single inducer can influence test-line judgments, even when two test-lines are presented, and can produce attractive and repulsive serial dependence biases when the item to report is revealed after it has disappeared.

Priming of probabilistic attentional templates

Attentional priming has a dominating influence on vision, speeding visual search, releasing items from crowding, reducing masking effects, and during free-choice, primed targets are chosen over unprimed ones. Many accounts postulate that templates stored in working memory control what we attend to and mediate the priming. But what is the nature of these templates (or representations)? Analyses of real-world visual scenes suggest that tuning templates to exact color or luminance values would be impractical since those can vary greatly because of changes in environmental circumstances and perceptual interpretation. Tuning templates to a range of the most probable values would be more efficient. Recent evidence does indeed suggest that the visual system represents such probability, gradually encoding statistical variation in the environment through repeated exposure to input statistics. This is consistent with evidence from neurophysiology and theoretical neuroscience as well as computational evidence of probabilistic representations in visual perception. I argue that such probabilistic representations are the unit of attentional priming and that priming of, say, a repeated single-color value simply involves priming of a distribution with no variance. This "priming of probability" view can be modelled within a Bayesian framework where priming provides contextual priors. Priming can therefore be thought of as learning of the underlying probability density function of the target or distractor sets in a given continuous task.

Statistical learning of distractor shape modulates attentional capture

Physically salient but task-irrelevant stimuli have high attentional priority, although observers are able to capitalize on statistical regularities in the environment to more efficiently ignore such stimuli. Physically salient distractors that more frequently appear in a particular location are less distracting when they appear in this high probability location. Likewise, colors and orientations that are frequently associated with distractors become preferentially ignored with learning. Such statistically learned distractor suppression has been examined with respect to the frequency of elementary features across trials, and less is known about how statistics concerning the composition of distractor features within a trial influence attention, particularly with respect to how orientations combine to form shapes. Color, orientation, and location are also represented very early in vision, whereas more complex features such as shape are represented further downstream in the visual system; it remains unclear whether statistically leaned distractor suppression can operate over such downstream visual representations. In the present study, we demonstrate attentional capture by physically salient, shape-defined distractors that is reduced in magnitude for a high probability shape. Our findings demonstrate that statistical learning can modulate attentional priority at least at the level of basic shapes and is not restricted to modulations of priority at the earliest stages of visual information processing tied to elementary features.

What kind of empirical evidence is needed for probabilistic mental representations? An example from visual perception

Recent accounts of perception and cognition propose that the brain represents information probabilistically. While this assumption is common, empirical support for such probabilistic representations in perception has recently been criticized. Here, we evaluate these criticisms and present an account based on a recently developed psychophysical methodology, Feature Distribution Learning (FDL), which provides promising evidence for probabilistic representations by avoiding these criticisms. The method uses priming and role-reversal effects in visual search. Observers' search times reveal the structure of perceptual representations, in which the probability distribution of distractor features is encoded. We explain how FDL results provide evidence for a stronger notion of representation that relies on structural correspondence between stimulus uncertainty and perceptual representations, rather than a mere co-variation between the two. Moreover, such an account allows us to demonstrate what kind of empirical evidence is needed to support probabilistic representations as posited in current probabilistic Bayesian theories of perception.

You see what you look for: Targets and distractors in visual search can cause opposing serial dependencies

Visual perception is, at any given moment, strongly influenced by its temporal context-what stimuli have recently been perceived and in what surroundings. We have previously shown that to-be-ignored items produce a bias upon subsequent perceptual decisions that acts in parallel with other biases induced by attended items. However, our previous investigations were confined to biases upon the perceived orientation of a visual search target, and it is unclear whether these biases influence perceptual decisions in a more general sense. Here, we test whether the biases from visual search targets and distractors affect the perceived orientation of a neutral test line, one that is neither a target nor a distractor. To do so, we asked participants to search for an oddly oriented line among distractors and report its location for a few trials and next presented a test line irrelevant to the search task. Participants were asked to report the orientation of the test line. Our results indicate that in tasks involving visual search, targets induce a positive bias upon a neutral test line if their orientations are similar, whereas distractors produce an attractive bias for similar test lines and a repulsive bias if the orientations of the test line and the average orientation of the distractors are far apart in feature space. In sum, our results show that both attentional role and proximity in feature space between previous and current stimuli determine the direction of biases in perceptual decisions.

Testing temporal integration of feature probability distributions using role-reversal effects in visual search

The visual system is sensitive to statistical properties of complex scenes and can encode feature probability distributions in detail. But does the brain use these statistics to build probabilistic models of the ever-changing visual input? To investigate this, we examined how observers temporally integrate two different orientation distributions from sequentially presented visual search trials. If the encoded probabilistic information is used in a Bayesian optimal way, observers should weigh more reliable information more strongly, such as feature distributions with low variance. We therefore manipulated the variance of the two feature distributions. Participants performed sequential odd-one-out visual search for an oddly oriented line among distractors. During successive learning trials, the distractor orientations were sampled from two different Gaussian distributions on alternating trials. Then, observers performed a 'test trial' where the orientations of the target and distractors were switched, allowing us to assess observer's internal representation of distractor distributions based on changes in response times. In three experiments we observed that observer's search times on test trials depended mainly on the very last learning trial, indicating a strong recency effect. Since temporal integration has been previously observed with this method, we conclude that when the input is unreliable, the visual system relies more on the most recent stimulus. This indicates that the visual system prefers to utilize sensory history when the statistical properties of the environment are relatively stable.

Keeping it real: Looking beyond capacity limits in visual cognition

Research within visual cognition has made tremendous strides in uncovering the basic operating characteristics of the visual system by reducing the complexity of natural vision to artificial but well-controlled experimental tasks and stimuli. This reductionist approach has for example been used to assess the basic limitations of visual attention, visual working memory (VWM) capacity, and the fidelity of visual long-term memory (VLTM). The assessment of these limits is usually made in a pure sense, irrespective of goals, actions, and priors. While it is important to map out the bottlenecks our visual system faces, we focus here on selected examples of how such limitations can be overcome. Recent findings suggest that during more natural tasks, capacity may be higher than reductionist research suggests and that separable systems subserve different actions, such as reaching and looking, which might provide important insights about how pure attentional or memory limitations could be circumvented. We also review evidence suggesting that the closer we get to naturalistic behavior, the more we encounter implicit learning mechanisms that operate "for free" and "on the fly." These mechanisms provide a surprisingly rich visual experience, which can support capacity-limited systems. We speculate whether natural tasks may yield different estimates of the limitations of VWM, VLTM, and attention, and propose that capacity measurements should also pass the real-world test within naturalistic frameworks. Our review highlights various approaches for this and suggests that our understanding of visual cognition will benefit from incorporating the complexities of real-world cognition in experimental approaches.

Optimizing perception: Attended and ignored stimuli create opposing perceptual biases

Humans have remarkable abilities to construct a stable visual world from continuously changing input. There is increasing evidence that momentary visual input blends with previous input to preserve perceptual continuity. Most studies have shown that such influences can be traced to characteristics of the attended object at a given moment. Little is known about the role of ignored stimuli in creating this continuity. This is important since while some input is selected for processing, other input must be actively ignored for efficient selection of the task-relevant stimuli. We asked whether attended targets and actively ignored distractor stimuli in an odd-one-out search task would bias observers' perception differently. Our observers searched for an oddly oriented line among distractors and were occasionally asked to report the orientation of the last visual search target they saw in an adjustment task. Our results show that at least two opposite biases from past stimuli influence current perception: A positive bias caused by serial dependence pulls perception of the target toward the previous target features, while a negative bias induced by the to-be-ignored distractor features pushes perception of the target away from the distractor distribution. Our results suggest that to-be-ignored items produce a perceptual bias that acts in parallel with other biases induced by attended items to optimize perception. Our results are the first to demonstrate how actively ignored information facilitates continuity in visual perception.

Structural and contextual priors affect visual search in children with and without autism

Bayesian predictive coding theories of autism spectrum disorder propose that impaired acquisition or a broader shape of prior probability distributions lies at the core of the condition. However, we still know very little about how probability distributions are learned and encoded by children, let alone children with autism. Here, we take advantage of a recently developed distribution learning paradigm to characterize how children with and without autism acquire information about probability distributions. Twenty-four autistic and 25-matched neurotypical children searched for an odd-one-out target among a set of distractor lines with orientations sampled from a Gaussian distribution repeated across multiple trials to allow for learning of the parameters (mean and variance) of the distribution. We could measure the width (variance) of the participant's encoded distribution by introducing a target-distractor role-reversal while varying the similarity between target and previous distractor mean. Both groups performed similarly on the visual search task and learned the distractor distribution to a similar extent. However, the variance learned was much broader than the one presented, consistent with less informative priors in children irrespective of autism diagnosis. These findings have important implications for Bayesian accounts of perception throughout development, and Bayesian accounts of autism specifically. LAY SUMMARY: Recent theories about the underlying cognitive mechanisms of autism propose that the way autistic individuals estimate variability or uncertainty in their perceptual environment may differ from how typical individuals do so. Children had to search an oddly tilted line in a set of lines pointing in different directions, and based on their response times we examined how they learned about the variability in a set of objects. We found that autistic children learn variability as well as typical children, but both groups learn with less precision than typical adults do on the same task.

Feature-Based Attentional Weighting and Re-weighting in the Absence of Visual Awareness

Visual attention evolved as an adaptive mechanism allowing us to cope with a rapidly changing environment. It enables the facilitated processing of relevant information, often automatically and governed by implicit motives. However, despite recent advances in understanding the relationship between consciousness and visual attention, the functional scope of unconscious attentional control is still under debate. Here, we present a novel masking paradigm in which volunteers were to distinguish between varying orientations of a briefly presented, masked grating stimulus. Combining signal detection theory and subjective measures of awareness, we show that performance on unaware trials was consistent with visual selection being weighted towards repeated orientations of Gabor patches and reallocated in response to a novel unconsciously processed orientation. This was particularly present in trials in which the prior feature was strongly weighted and only if the novel feature was invisible. Thus, our results provide evidence that invisible orientation stimuli can trigger the reallocation of history-guided visual selection weights.

Encoding perceptual ensembles during visual search in peripheral vision

Observers can learn complex statistical properties of visual ensembles, such as their probability distributions. Even though ensemble encoding is considered critical for peripheral vision, whether observers learn such distributions in the periphery has not been studied. Here, we used a visual search task to investigate how the shape of distractor distributions influences search performance and ensemble encoding in peripheral and central vision. Observers looked for an oddly oriented bar among distractors taken from either uniform or Gaussian orientation distributions with the same mean and range. The search arrays were either presented in the foveal or peripheral visual fields. The repetition and role reversal effects on search times revealed observers' internal model of distractor distributions. Our results showed that the shape of the distractor distribution influenced search times only in foveal, but not in peripheral search. However, role reversal effects revealed that the shape of the distractor distribution could be encoded peripherally depending on the interitem spacing in the search array. Our results suggest that, although peripheral vision might rely heavily on summary statistical representations of feature distributions, it can also encode information about the distributions themselves.

Distractor ignoring: strategies, learning, and passive filtering

Our sensory environments contain more information than we can processes and successful behaviors require the ability to separate task-relevant information from task-irrelevant information. While much research on attention has focused on the mechanisms that result in selection of desired information, much less is known about how distracting information is ignored. Here we describe evidence that strategic, learned, and passive information can all contribute to better distractor ignoring. The evidence suggests that there are multiple ways in which distractor ignoring is supported that may be different than those of target selection. Future work will need to identify the mechanisms by which each source of information adjusts attentional priority such that irrelevant information is better ignored.

Learned feature variance is encoded in the target template and drives visual search

Real world visual search targets are frequently imperfect perceptual matches to our internal target templates. For example, the same friend on different occasions is likely to wear different clothes, hairstyles, and accessories, but some of these may be more likely to vary than others. The ability to deal with template-to-target variability is important to visual search in natural environments, but we know relatively little about how feature variability is handled by the attentional system. In these studies, we test the hypothesis that top-down attentional biases are sensitive to the variance of target feature dimensions over time and prioritize information from less-variable dimensions. On each trial, subjects were shown a target cue composed of colored dots moving in a specific direction followed by a working memory probe (30%) or visual search display (70%). Critically, the target features in the visual search display differed from the cue, with one feature drawn from a distribution narrowly centered over the cued feature (low-variance dimension), and the other sampled from a broader distribution (high-variance dimension). The results demonstrate that subjects used knowledge of the likely cue-to-target variance to set template precision and bias attentional selection. Moreover, an individual's working memory precision for each feature predicted search performance. Our results suggest that observers are sensitive to the variance of feature dimensions within a target and use this information to weight mechanisms of attentional selection.

Continuous to discrete: Ensemble-based segmentation in the perception of multiple feature conjunctions

Although objects around us vary in a number of continuous dimensions (color, size, orientation, etc.), we tend to perceive the objects using more discrete, categorical descriptions (e.g., berries and leaves). Previously, we described how continuous ensemble statistics of simple features are transformed into categorical classes: The visual system tests whether the feature distribution has one or several peaks, each representing a likely "category". Here, we tested the mechanism of segmentation for more complex conjunctions of features. Observers discriminated between two textures filled with lines of various lengths and orientations, which had same distributions between the textures, but opposite directions of correlations. Critically, feature distributions could be "segmentable" (only extreme feature values and a large gap between them) or "non-segmentable" (both extreme and middle values with smooth transition are present). Segmentable displays yielded steeper psychometric functions indicating better discrimination (Experiment 1). The effect of segmentability arises early in visual processing (Experiment 2) and is likely to be provided by global sampling of the entire field (Experiment 3). Also, rapid segmentation requires both feature dimensions having a "segmentable" distribution supporting division of the textures into categorical classes of conjunctions. We propose that observers select items from one side (peak) of one dimension and sample mean differences along a second dimension within the selected subset. In this scenario, subset selection is a limiting factor (Experiment 4) of texture discrimination. Yet, segmentability provided by the sharp feature distributions seems to facilitate both subset selection and mean comparison.