The case for implicit category learning

Testing formal cognitive models of classification and old-new recognition in a real-world high-dimensional category domain

Categorization and old-new recognition memory are closely linked topics in the cognitive-psychology literature and there have been extensive past efforts at developing unified formal modeling accounts of these fundamental psychological processes. However, the existing formal-modeling literature has almost exclusively used small sets of simplified stimuli and artificial category structures. The present work extends this literature by collecting both categorization and old-new recognition judgments on a large set of high-dimensional stimuli that form real-world category structures: namely, a set of 540 images of rocks belonging to the geologically-defined categories igneous, metamorphic and sedimentary. Participants first engaged in a learning phase in which they classified large sets of training instances into these real-world categories. This was followed by a test phase in which they classified both training and novel transfer items into the learned categories and also judged whether each item was old or new. We attempted to model both the classification and recognition test data at the level of individual items. Ultimately, the categorization data were well fit by both an exemplar and clustering model, but not by a prototype model. Only the exemplar model was able to provide a reasonable first-order account of the old-new recognition data; however, the standard version of the model failed to capture the variability in hit rates within the class of old-training items themselves. An extended hybrid-similarity version of the exemplar model that made allowance for boosts in self-similarity due to matching distinctive features yielded much improved accounts of the old-new recognition data. The study is among the first to test cognitive-process models on their ability to account quantitatively for old-new recognition of real-world, high-dimensional stimuli at the level of individual items.

Deep learning can be used to train naïve, nonprofessional observers to detect diagnostic visual patterns of certain cancers in mammograms: a proof-of-principle study

The scientific, clinical, and pedagogical significance of devising methodologies to train nonprofessional subjects to recognize diagnostic visual patterns in medical images has been broadly recognized. However, systematic approaches to doing so remain poorly established. Using mammography as an exemplar case, we use a series of experiments to demonstrate that deep learning (DL) techniques can, in principle, be used to train naïve subjects to reliably detect certain diagnostic visual patterns of cancer in medical images. In the main experiment, subjects were required to learn to detect statistical visual patterns diagnostic of cancer in mammograms using only the mammograms and feedback provided following the subjects' response. We found not only that the subjects learned to perform the task at statistically significant levels, but also that their eye movements related to image scrutiny changed in a learning-dependent fashion. Two additional, smaller exploratory experiments suggested that allowing subjects to re-examine the mammogram in light of various items of diagnostic information may help further improve DL of the diagnostic patterns. Finally, a fourth small, exploratory experiment suggested that the image information learned was similar across subjects. Together, these results prove the principle that DL methodologies can be used to train nonprofessional subjects to reliably perform those aspects of medical image perception tasks that depend on visual pattern recognition expertise.

The effect of encoding conditions on learning in the prototype distortion task

The prototype distortion task demonstrates that it is possible to learn about a category of physically similar stimuli through mere observation. However, there have been few attempts to test whether different encoding conditions affect learning in this task. This study compared prototypicality gradients produced under incidental learning conditions in which participants performed a visual search task, with those produced under intentional learning conditions in which participants were required to memorize the stimuli. Experiment 1 showed that similar prototypicality gradients could be obtained for category endorsement and familiarity ratings, but also found (weaker) prototypicality gradients in the absence of exposure. In Experiments 2 and 3, memorization was found to strengthen prototypicality gradients in familiarity ratings in comparison to visual search, but there were no group differences in participants' ability to discriminate between novel and presented exemplars. Although the Search groups in Experiments 2 and 3 produced prototypicality gradients, they were no different in magnitude to those produced in the absence of stimulus exposure in Experiment 1, suggesting that incidental learning during visual search was not conducive to producing prototypicality gradients. This study suggests that learning in the prototype distortion task is not implicit in the sense of resulting automatically from exposure, is affected by the nature of encoding, and should be considered in light of potential learning-at-test effects.

Category learning in Alzheimer's disease and normal cognitive aging depends on initial experience of feature variability

Semantic category learning is dependent upon several factors, including the nature of the learning task, as well as individual differences in the quality and heterogeneity of exemplars that an individual encounters during learning. We trained healthy older adults (n=39) and individuals with a diagnosis of Alzheimer's disease or Mild Cognitive Impairment (n=44) to recognize instances of a fictitious animal, a "crutter". Each stimulus item contained 10 visual features (e.g., color, tail shape) which took one of two values for each feature (e.g., yellow/red, curly/straight tails). Participants were presented with a series of items (learning phase) and were either told the items belonged to a semantic category (explicit condition) or were told to think about the appearance of the items (implicit condition). Half of participants saw learning items with higher similarity to an unseen prototype (high typicality learning set), and thus lower between-item variability in their constituent features; the other half learned from items with lower typicality (low typicality learning set) and higher between-item feature variability. After the learning phase, participants were presented with test items one at a time that varied in the number of typical features from 0 (antitype) to 10 (prototype). We examined between-subjects factors of learning set (lower or higher typicality), instruction type (explicit or implicit), and group (patients vs. elderly control). Learning in controls was aided by higher learning set typicality: while controls in both learning set groups demonstrated significant learning, those exposed to a high-typicality learning set appeared to develop a prototype that helped guide their category membership judgments. Overall, patients demonstrated more difficulty with category learning than elderly controls. Patients exposed to the higher-typicality learning set were sensitive to the typical features of the category and discriminated between the most and least typical test items, although less reliably than controls. In contrast, patients exposed to the low-typicality learning set showed no evidence of learning. Analysis of structural imaging data indicated a positive association between left hippocampal grey matter density in elderly controls but a negative association in the patient group, suggesting differential reliance on hippocampal-mediated learning. Contrary to hypotheses, learning did not differ between explicit and implicit conditions for either group. Results demonstrate that category learning is improved when learning materials are highly similar to the prototype.

Invariance detection in the brain: revealed in a stepwise category induction task

A critical sub-process of category learning is detecting the invariance between categorical members. To examine brain activation associated with invariance detection at different steps of category learning, a stepwise category induction task was used in the present study. Within each trial, three stimuli were displayed sequentially, and participants were asked to learn the target category corresponding to the invariance among stimuli. Results revealed that invariance detection activated the fronto-parietal network. However, the frontal and parietal cortices functioned differently throughout the different steps of invariance detection. The left middle frontal gyrus (BA 9) was highly activated in both steps of invariance detection, but the posterior parietal regions, especially the right superior parietal lobule (BA 7), were more active in the final step of invariance detection, reflecting increased attention to the completion of category learning and the preparation for a subsequent response. Furthermore, a psychophysiological interaction analysis (PPI) revealed increased connectivity between the left middle frontal gyrus and the bilateral parietal cortex during the final step of invariance detection. Overall, the present findings imply the necessary role of the fronto-parietal network in variance detection.

Predicting and interpreting identification errors in military vehicle training using multidimensional scaling

We compared methods for predicting and understanding the source of confusion errors during military vehicle identification training. Participants completed training to identify main battle tanks. They also completed card-sorting and similarity-rating tasks to express their mental representation of resemblance across the set of training items. We expected participants to selectively attend to a subset of vehicle features during these tasks, and we hypothesised that we could predict identification confusion errors based on the outcomes of the card-sort and similarity-rating tasks. Based on card-sorting results, we were able to predict about 45% of observed identification confusions. Based on multidimensional scaling of the similarity-rating data, we could predict more than 80% of identification confusions. These methods also enabled us to infer the dimensions receiving significant attention from each participant. This understanding of mental representation may be crucial in creating personalised training that directs attention to features that are critical for accurate identification.

PRACTITIONER SUMMARY

Participants completed military vehicle identification training and testing, along with card-sorting and similarity-rating tasks. The data enabled us to predict up to 84% of identification confusion errors and to understand the mental representation underlying these errors. These methods have potential to improve training and reduce identification errors leading to fratricide.

Experience produces the atypicality bias in object perception

When a morph face is produced with equal physical contributions from a typical parent face and an atypical parent face, the morph is judged to be more similar to the atypical parent. This discontinuity between physical and perceptual distance relationships, called the "atypicality bias" (Tanaka et al 1998, Cognition 68 199-220), has also been demonstrated with non-face objects (birds and cars; Tanaka and Corneille 2007 Perception & Psychophysics 69 619-627). We tested whether the atypicality bias can be induced for a novel set of artificial objects. Two categories of "blob" stimuli were generated, each composed of typical and atypical members. Morphs averaged from typical and atypical parent exemplars were used to test the presence of an atypicality bias before and after participants were familiarized with blob items. In experiment 1, participants were trained to discriminate between the two blob categories. An atypicality bias was evident after, but not prior to, category training. In experiment 2, participants rated the pleasantness of the blobs instead of learning to categorize them; an atypicality bias was present only after the ratings task. This finding suggests that relatively passive exposure to exemplars is sufficient to influence perceptions of similarity, and that the atypicality bias is a manifestation of this influence.

Perceptual fluency can be used as a cue for categorization decisions

Learning in the prototype distortion task is thought to involve perceptual learning in which category members experience an enhanced visual response (Ashby & Maddox. Annual Review of Psychology, 56, 149-178, 2005). This response likely leads to more-efficient processing, which in turn may result in a feeling of perceptual fluency for category members. We examined the perceptual-fluency hypothesis by manipulating fluency independently from category membership. We predicted that when perceptual fluency was induced using subliminal priming, this fluency would be misattributed to category membership and would affect categorization decisions. In a prototype distortion task, the participants were more likely to judge stimuli that were not members of the category as category members when the nonmembers were made perceptually fluent with a matching subliminal prime. This result suggests that perceptual fluency can be used as a cue during some categorization decisions. In addition, the results provided converging evidence that some types of categorization are based on perceptual learning.

Activation in the neural network responsible for categorization and recognition reflects parameter changes

According to various influential formal models of cognition, perceptual categorization and old-new recognition recruit the same memory system. By contrast, the prevailing view in the cognitive neuroscience literature is that separate neural systems mediate perceptual categorization and recognition. A direct form of evidence is that separate brain regions are activated when observers engage in categorization and recognition tasks involving the same types of stimuli. However, even if the same memory-based comparison processes underlie categorization and recognition, one would not expect to see identical patterns of brain activity across the tasks; the reason is that observers would adjust parameter settings (e.g., vary criterion settings) across the tasks to satisfy the different task goals. In this fMRI study, we conducted categorization and recognition tasks in which stimulus conditions were held constant, and in which observers were induced to vary hypothesized parameter settings across conditions. A formal exemplar model was fitted to the data to track the changes in parameters to help interpret the fMRI results. We observed systematic effects of changes in parameters on patterns of brain activity, which were interpretable in terms of differing forms of evidence accumulation that resulted from the changed parameter settings. After controlling for stimulus and parameter-related differences, we found little evidence that categorization and recognition recruit separate memory systems.

Category learning in the brain

The ability to group items and events into functional categories is a fundamental characteristic of sophisticated thought. It is subserved by plasticity in many neural systems, including neocortical regions (sensory, prefrontal, parietal, and motor cortex), the medial temporal lobe, the basal ganglia, and midbrain dopaminergic systems. These systems interact during category learning. Corticostriatal loops may mediate recursive, bootstrapping interactions between fast reward-gated plasticity in the basal ganglia and slow reward-shaded plasticity in the cortex. This can provide a balance between acquisition of details of experiences and generalization across them. Interactions between the corticostriatal loops can integrate perceptual, response, and feedback-related aspects of the task and mediate the shift from novice to skilled performance. The basal ganglia and medial temporal lobe interact competitively or cooperatively, depending on the demands of the learning task.

Re-evaluating dissociations between implicit and explicit category learning: an event-related fMRI study

Recent fMRI studies have found that distinct neural systems may mediate perceptual category learning under implicit and explicit learning conditions. In these previous studies, however, different stimulus-encoding processes may have been associated with implicit versus explicit learning. The present design was aimed at decoupling the influence of these factors on the recruitment of alternate neural systems. Consistent with previous reports, following incidental learning in a dot-pattern classification task, participants showed decreased neural activity in occipital visual cortex (extrastriate region V3, BA 19) in response to novel exemplars of a studied category compared to members of a foil category, but did not show this decreased neural activity following explicit learning. Crucially, however, our results show that this pattern was primarily modulated by aspects of the stimulus-encoding instructions provided at the time of study. In particular, when participants in an implicit learning condition were encouraged to evaluate the overall shape and configuration of the stimuli during study, we failed to find the pattern of brain activity that has been taken to be a signature of implicit learning, suggesting that activity in this area does not uniquely reflect implicit memory for perceptual categories but instead may reflect aspects of processing or perceptual encoding strategies.

Medial temporal lobe involvement in an implicit memory task: evidence of collaborating implicit and explicit memory systems from FMRI and Alzheimer's disease

We used a prototype extraction task to assess implicit learning of a meaningful novel visual category. Cortical activation was monitored in young adults with functional magnetic resonance imaging. We observed occipital deactivation at test consistent with perceptually based implicit learning, and lateral temporal cortex deactivation reflecting implicit acquisition of the category's semantic nature. Medial temporal lobe (MTL) activation during exposure and test suggested involvement of explicit memory as well. Behavioral performance of Alzheimer's disease (AD) patients and healthy seniors was also assessed, and AD performance was correlated with gray matter volume using voxel-based morphometry. AD patients showed learning, consistent with preserved implicit memory, and confirming that AD patients' implicit memory is not limited to abstract patterns. However, patients were somewhat impaired relative to healthy seniors. Occipital and lateral temporal cortical volume correlated with successful AD patient performance, and thus overlapped with young adults' areas of deactivation. Patients' severe MTL atrophy precluded involvement of this region. AD patients thus appear to engage a cortically based implicit memory mechanism, whereas their relative deficit on this task may reflect their MTL disease. These findings suggest that implicit and explicit memory systems collaborate in neurologically intact individuals performing an ostensibly implicit memory task.