Modeling recognition memory using the similarity structure of natural input

The natural input memory (NIM) model is a new model for recognition memory that operates on natural visual input. A biologically informed perceptual preprocessing method takes local samples (eye fixations) from a natural image and translates these into a feature-vector representation. During recognition, the model compares incoming preprocessed natural input to stored representations. By complementing the recognition memory process with a perceptual front end, the NIM model is able to make predictions about memorability based directly on individual natural stimuli. We demonstrate that the NIM model is able to simulate experimentally obtained similarity ratings and recognition memory for individual stimuli (i.e., face images).     

Uncovering mental representations with Markov chain Monte Carlo

A key challenge for cognitive psychology is the investigation of mental representations, such as object categories, subjective probabilities, choice utilities, and memory traces. In many cases, these representations can be expressed as a non-negative function defined over a set of objects. We present a behavioral method for estimating these functions. Our approach uses people as components of a Markov chain Monte Carlo (MCMC) algorithm, a sophisticated sampling method originally developed in statistical physics. Experiments 1 and 2 verified the MCMC method by training participants on various category structures and then recovering those structures. Experiment 3 demonstrated that the MCMC method can be used estimate the structures of the real-world animal shape categories of giraffes, horses, dogs, and cats. Experiment 4 combined the MCMC method with multidimensional scaling to demonstrate how different accounts of the structure of categories, such as prototype and exemplar models, can be tested, producing samples from the categories of apples, oranges, and grapes.     

What Is the Model in Model-Based Planning?

Flexibility is one of the hallmarks of human problem-solving. In everyday life, people adapt to changes in common tasks with little to no additional training. Much of the existing work on flexibility in human problem-solving has focused on how people adapt to tasks in new domains by drawing on solutions from previously learned domains. In real-world tasks, however, humans must generalize across a wide range of within-domain variation. In this work we argue that representational abstraction plays an important role in such within-domain generalization. We then explore the nature of this representational abstraction in realistically complex tasks like video games by demonstrating how the same model-based planning framework produces distinct generalization behaviors under different classes of task representation. Finally, we compare the behavior of agents with these task representations to humans in a series of novel grid-based video game tasks. Our results provide evidence for the claim that within-domain flexibility in humans derives from task representations composed of propositional rules written in terms of objects and relational categories.     

Natural scene categorization with minimal attention: evidence from negative priming

Research has demonstrated that objects in natural scenes are categorized without the deployment of attention. However, in these types of studies, participants were required to directly respond to peripherally presented scenes, which might lead some participants to move their attention. If this is the case, the above conclusion concerning natural scenes may not be valid. We investigated this issue by using a negative priming (NP) paradigm in which participants did not directly respond to peripheral stimuli. Our results showed NP effect from ignored stimuli in natural scene categorization, but neither in letter discrimination (Experiment 1) nor in line-drawing categorization (Experiment 2). In addition, NP effects were observed even when probe stimuli were words (Experiments 3A and 3B). These findings suggest that people can categorize objects in natural scenes with minimal attention, that this process is specific to natural scenes, and that it is based on the semantic information of the images.     

Extracting Low-Dimensional Psychological Representations from Convolutional Neural Networks

Convolutional neural networks (CNNs) are increasingly widely used in psychology and neuroscience to predict how human minds and brains respond to visual images. Typically, CNNs represent these images using thousands of features that are learned through extensive training on image datasets. This raises a question: How many of these features are really needed to model human behavior? Here, we attempt to estimate the number of dimensions in CNN representations that are required to capture human psychological representations in two ways: (1) directly, using human similarity judgments and (2) indirectly, in the context of categorization. In both cases, we find that low-dimensional projections of CNN representations are sufficient to predict human behavior. We show that these low-dimensional representations can be easily interpreted, providing further insight into how people represent visual information. A series of control studies indicate that these findings are not due to the size of the dataset we used and may be due to a high level of redundancy in the features appearing in CNN representations.     

Regular Minimality and Thurstonian-type modeling

A Thurstonian-type model for pairwise comparisons is any model in which the response (e.g., “they are the same” or “they are different”) to two stimuli being compared depends, deterministically or probabilistically, on the realizations of two randomly varying representations (perceptual images) of these stimuli. The two perceptual images in such a model may be stochastically interdependent but each has to be selectively dependent on its stimulus. It has been previously shown that all possible discrimination probability functions for same–different comparisons can be generated by Thurstonian-type models of the simplest variety, with independent percepts and deterministic decision rules. It has also been shown, however, that a broad class of Thurstonian-type models, called “well-behaved” (and including, e.g., models with multivariate normal perceptual representations whose parameters are smooth functions of stimuli) cannot simultaneously account for two empirically plausible properties of same–different comparisons, Regular Minimality (which essentially says that “being least discriminable from” is a symmetric relation) and nonconstancy of the minima of discrimination probabilities (the fact that different pairs of least discriminable stimuli are discriminated with different probabilities). These results have been obtained for stimulus spaces represented by regions of Euclidean spaces. In this paper, the impossibility for well-behaved Thurstonian-type models to simultaneously account for Regular Minimality and nonconstancy of minima is established for a much broader notion of well-behavedness applied to a much broader class of stimulus spaces (any Hausdorff arc-connected ones). The universality of Thurstonian-type models with independent perceptual images and deterministic decision rules is shown (by a simpler proof than before) to hold for arbitrary stimulus spaces.     

Electrophysiological evidence for a shared representational medium for visual images and visual percepts

Does mental imagery involve the activation of representations in the visual system? Systematic effects of imagery on visual signal detection performance have been used to argue that imagery and the perceptual processing of stimuli interact at some common locus of activity (Farah, 1985). However, such a result is neutral with respect to the question of whether the interaction occurs during modality-specific visual processing of the stimulus. If imagery affects stimulus processing at early, modality-specific stages of stimulus representation, this implies that the shared stimulus representations are visual, whereas if imagery affects stimulus processing only at later, amodal stages of stimulus representation, this implies that imagery involves more abstract, postvisual stimulus representations. To distinguish between these two possibilities, we repeated the earlier imagery-perception interaction experiment while recording event-related potentials (ERPs) to stimuli from 16 scalp electrodes. By observing the time course and scalp distribution of the effect of imagery on the ERP to stimuli, we can put constraints on the locus of the shared representations for imagery and perception. An effect of imagery was seen within 200 ms following stimulus presentation, at the latency of the first negative component of the visual ERP, localized at the occipital and posterior temporal regions of the scalp, that is, directly over visual cortex. This finding provides support for the claim that mental images interact with percepts in the visual system proper and hence that mental images are themselves visual representations.     

Visualising mental representations: A primer on noise-based reverse correlation in social psychology

With the introduction of the psychophysical method of reverse correlation, a holy grail of social psychology appears to be within reach – visualising mental representations. Reverse correlation is a data-driven method that yields visual proxies of mental representations, based on judgements of randomly varying stimuli. This review is a primer to an influential reverse correlation approach in which stimuli vary by applying random noise to the pixels of images. Our review suggests that the technique is an invaluable tool in the investigation of social perception (e.g., in the perception of race, gender and personality traits), with ample potential applications. However, it is unclear how these visual proxies are best interpreted. Building on advances in cognitive neuroscience, we suggest that these proxies are visual reflections of the internal representations that determine how social stimuli are perceived. In addition, we provide a tutorial on how to perform reverse correlation experiments using R.     

Effects of stimulus duration and choice delay on visual categorization in pigeons

We [Lazareva, O. F., Freiburger, K. L., & Wasserman, E. A. (2004). Pigeons concurrently categorize photographs at both basic and superordinate levels. Psychonomic Bulletin and Review, 11, 1111-1117] previously trained four pigeons to classify color photographs into their basic-level categories (cars, chairs, flowers, or people) or into their superordinate-level categories (natural or artificial). Here, we found that brief stimulus durations had the most detrimental effect on the basic-level discrimination of natural stimuli by the same pigeons. Increasing the delay between stimulus presentation and choice responding had greater detrimental effect on the basic-level discrimination than the superordinate-level discrimination. These results suggest that basic-level discriminations required longer stimulus durations and were more subject to forgetting than were superordinate-level discriminations. Additionally, categorization of natural stimuli required longer stimulus durations than categorization of artificial stimuli, but only at the basic level. Together, these findings suggest that basic-level categorization may not always be superior to superordinate-level categorization and provide additional evidence of a dissociation between natural and artificial stimuli in pigeons’ categorization.     

Pigeons concurrently categorize photographs at both basic and superordinate levels

We studied categorization in pigeons, using carefully controlled photographs. Within daily sessions, 4 pigeons had to classify each of 32 photographs into either its proper basic-level category (cars, chairs, flowers, or people; four-key forced choice procedure) or its proper superordinate-level category (natural or artificial; two-key forced choice procedure). The pigeons successfully classified the same stimuli at both levels. Overall, the pigeons learned the basic discrimination more quickly than the superordinate discrimination, but this difference was reliable only for artificial stimuli (cars and chairs), not for natural stimuli (flowers and people). The pigeons also exhibited reliable discrimination transfer to novel photographs, attesting to the open-endedness of these basic and superordinate categories.     

Spatial categories and the estimation of location

Four experiments are reported in which people organize a space hierarchically when they estimate particular locations in that space. Earlier work showed that people subdivide circles into quadrants bounded at the vertical and horizontal axes, biasing their estimates towards prototypical diagonal locations within those spatial categories (Psychological Review 98 (1991) 352). In this work Huttenlocher, Hedges, and Duncan showed that the use of such spatial categories can increase the accuracy of estimation of inexactly represented locations. The stimulus locations we examined were uniformly distributed across the circle. In the present study we explore whether variation in the distribution of locations affects how the circle is categorized. Other things being equal, categories that capture high density regions in a stimulus space should contribute most to accuracy of estimation. However, precision of boundaries is also important to accuracy; with imprecise boundaries stimuli may be misclassified, leading to large errors in estimation. We found that people use the same spatial categories regardless of the distribution of the locations. We argue that this spatial organization nevertheless can maximize the accuracy of estimates because vertical and horizontal category boundaries are the most exact, minimizing misclassification of stimuli.     

Ordinality and novel sequence learning in jackdaws

A hallmark of higher cognition is the flexible use of information. This requires an abstract representation of the information. In sequence learning, ordinal position knowledge is seen as a more versatile representation when compared to chaining. Here, we assessed which of these mental representations is the most natural and most dominant in jackdaws. Two jackdaws (Corvus monedula) were trained on 14 separate three-item sequences (triplets), made up of abstract images. On each trial, the three items of one triplet were presented in fixed order. The images represented either the first, second or third ordinal position. Test stimuli consisted of the three images and a distractor image that was chosen randomly from the remaining sequences. We rewarded pecking in the correct order to the images belonging to the same sequence. The most common error the birds made was to peck at a distractor item from the same ordinal position. To look at how versatile the jackdaws' ordinal knowledge was, we replaced a familiar item with a novel item in some sequences. We then created novel sequences with these items, which the birds completed correctly. It appears, then, that jackdaws have a concept of ordinal position.     

Analysis of the conceptual structure of natural and artifactual categories in Alzheimer's disease patients

This work has two goals, the first one is to study in detail the conceptual structure of some natural and artifactual categories, by using multidimensional scaling (MDS). According to our theoretical approximation, conceptual structure is composed of semantic features, and these features have different entities and nature. As an index of relevance in mental representation of semantic categories, we took the frequency of production of semantic features on a free verbal production task. Our second goal was to compare the structure of conceptual representations in two populations, healthy elderly and Alzheimer's disease (AD) patients, assuming that conceptual representation is impaired in this last population. We assume that the impairment in the conceptual representations of AD patients will show, when the disease reaches a determined level, loosing specific features. This impairment will have serious consequences in the whole conceptual structure.     

Two spatial memories are not better than one: evidence of exclusivity in memory for object location

This paper studies the dynamics of attempting to access two spatial memories simultaneously and its implications for the accuracy of recall. Experiment 1 demonstrates in a range of conditions that two cues pointing to different experiences of the same object location produce little or no higher recall than that observed with a single cue. Experiment 2 confirms this finding in a within-subject design where both cues have previously elicited recall. Experiment 3 shows that these findings are only consistent with a model in which two representations of the same object location are mutually exclusive at both encoding and retrieval, and inconsistent with models that assume information from both representations is available. We propose that these representations quantify directionally specific judgments of location relative to specific anchor points in the stimulus; a format that precludes the parallel processing of like representations. Finally, we consider the apparent paradox of how such representations might contribute to the acquisition of spatial knowledge from multiple experiences of the same stimuli.     

Discrimination and categorization of photographs of natural objects by chimpanzees (Pan troglodytes)

Two experiments assessed the ability of four adult female chimpanzees to categorize natural objects. Chimpanzees were initially trained to match different color photographs of familiar objects from four possible categories. In training, all the comparison stimuli were from the same category in one condition, and from different categories in another condition. For all subjects, training performance was consistently better for the "different category" than for the "same category" trials. Probe trials were shown after training. In probe trials, the sample and positive comparison stimuli were different items from the same category, and the foils were selected from among the three other test categories. Individual performance was above chance in probe trials, suggesting that categorization by chimpanzees may transcend perceptual resemblance. These results were later replicated with novel stimulus items from the same four categories (experiment 2). Altogether, this research demonstrates that chimpanzees grouped perceptually different exemplars within the same category, and further suggests that these animals formed conceptual representations of the categories. Accepted after revision: 10 August 2001 Electronic Publication     

Similarity Judgment Within and Across Categories: A Comprehensive Model Comparison

Similarity is one of the most important relations humans perceive, arguably subserving category learning and categorization, generalization and discrimination, judgment and decision making, and other cognitive functions. Researchers have proposed a wide range of representations and metrics that could be at play in similarity judgment, yet have not comprehensively compared the power of these representations and metrics for predicting similarity within and across different semantic categories. We performed such a comparison by pairing nine prominent vector semantic representations with seven established similarity metrics that could operate on these representations, as well as supervised methods for dimensional weighting in the similarity function. This approach yields a factorial model structure with 126 distinct representation-metric pairs, which we tested on a novel dataset of similarity judgments between pairs of cohyponymic words in eight categories. We found that cosine similarity and Pearson correlation were the overall best performing unweighted similarity functions, and that word vectors derived from free association norms often outperformed word vectors derived from text (including those specialized for similarity). Importantly, models that used human similarity judgments to learn category-specific weights on dimensions yielded substantially better predictions than all unweighted approaches across all types of similarity functions and representations, although dimension weights did not generalize well across semantic categories, suggesting strong category context effects in similarity judgment. We discuss implications of these results for cognitive modeling and natural language processing, as well as for theories of the representations and metrics involved in similarity.     

特征概率影响多维和少维类别的分类学习和特征学习

探讨了特征概率对多维和少维类别的分类学习和特征学习的效果及策略的影响。结果表明高特征概率条件下,多维比少维类别的分类学习更容易,而且学到更多的特征知识,多维条件下人们更倾向于整体性加工策略,而少维条件下人们倾向于分析性加工策略。低特征概率条件下,多维比少维类别的分类学习和特征学习都困难,且两种条件下人们都倾向于采取分析性加工策略。     

A category-overshadowing effect in pigeons: support for the Common Elements Model of object categorization learning

A model proposing error-driven learning of associations between representations of stimulus properties and responses can account for many findings in the literature on object categorization by nonhuman animals. Furthermore, the model generates predictions that have been confirmed in both pigeons and people, suggesting that these learning processes are widespread across distantly related species. The present work reports evidence of a category-overshadowing effect in pigeons' categorization of natural objects, a novel behavioral phenomenon predicted by the model. Object categorization learning was impaired when a second category of objects provided redundant information about correct responses. The same impairment was not observed when single objects provided redundant information, but the category to which they belonged was uninformative, suggesting that this effect is different from simple overshadowing, arising from competition among stimulus categories rather than individual stimuli during learning.     

Vision as Bayesian inference: analysis by synthesis?

We argue that the study of human vision should be aimed at determining how humans perform natural tasks with natural images. Attempts to understand the phenomenology of vision from artificial stimuli, although worthwhile as a starting point, can lead to faulty generalizations about visual systems, because of the enormous complexity of natural images. Dealing with this complexity is daunting, but Bayesian inference on structured probability distributions offers the ability to design theories of vision that can deal with the complexity of natural images, and that use 'analysis by synthesis' strategies with intriguing similarities to the brain. We examine these strategies using recent examples from computer vision, and outline some important implications for cognitive science.     

Représentations des compétences,une approche psychosociale

This paper presents studies conducted within the framework of social psychological theories and bearing on representations of competences (RC). The purpose is to demonstrate the social anchoring of such representations. We describe several processes that vary according to the type of occupation, organizational rationalities and intergroup relations. Study 1 shows that trainers and psychologists differ in the way they conceive the competences of unemployed people. Study 2 suggests that RC are tied with status and power games within organizations. Study set 3 shows that variations in representations of ingroup/outgroup competences can be explained by intergroup relations and identity phenomena. Theoretical and practical implications in the area of work and organizational psychology are discussed.