The Neurodynamics of Cognition: A Tutorial on Computational Cognitive Neuroscience

Computational Cognitive Neuroscience (CCN) is a new field that lies at the intersection of computational neuroscience, machine learning, and neural network theory (i.e., connectionism). The ideal CCN model should not make any assumptions that are known to contradict the current neuroscience literature and at the same time provide good accounts of behavior and at least some neuroscience data (e.g., single-neuron activity, fMRI data). Furthermore, once set, the architecture of the CCN network and the models of each individual unit should remain fixed throughout all applications. Because of the greater weight they place on biological accuracy, CCN models differ substantially from traditional neural network models in how each individual unit is modeled, how learning is modeled, and how behavior is generated from the network. A variety of CCN solutions to these three problems are described. A real example of this approach is described, and some advantages and limitations of the CCN approach are discussed.     

Grounding co-occurrence: Identifying features in a lexical co-occurrence model of semantic memory

Lexical co-occurrence models of semantic memory represent word meaning by vectors in a high-dimensional space. These vectors are derived from word usage, as found in a large corpus of written text. Typically, these models are fully automated, an advantage over models that represent semantics that are based on human judgments (e.g., feature-based models). A common criticism of co-occurrence models is that the representations are not grounded: Concepts exist only relative to each other in the space produced by the model. It has been claimed that feature-based models offer an advantage in this regard. In this article, we take a step toward grounding a cooccurrence model. A feed-forward neural network is trained using back propagation to provide a mapping from co-occurrence vectors to feature norms collected from subjects. We show that this network is able to retrieve the features of a concept from its co-occurrence vector with high accuracy and is able to generalize this ability to produce an appropriate list of features from the co-occurrence vector of a novel concept.     

Hierarchical growing grid networks for skeleton based action recognition

In this paper, a novel cognitive architecture for action recognition is developed by applying layers of growing grid neural networks. Using these layers makes the system capable of automatically arranging its representational structure. In addition to the expansion of the neural map during the growth phase, the system is provided with a prior knowledge of the input space, which increases the processing speed of the learning phase. Apart from two layers of growing grid networks the architecture is composed of a preprocessing layer, an ordered vector representation layer and a one-layer supervised neural network. These layers are designed to solve the action recognition problem. The first-layer growing grid receives the input data of human actions and the neural map generates an action pattern vector representing each action sequence by connecting the elicited activation of the trained map. The pattern vectors are then sent to the ordered vector representation layer to build the time-invariant input vectors of key activations for the second-layer growing grid. The second-layer growing grid categorizes the input vectors to the corresponding action clusters/sub-clusters and finally the one-layer supervised neural network labels the shaped clusters with action labels. Three experiments using different datasets of actions show that the system is capable of learning to categorize the actions quickly and efficiently. The performance of the growing grid architecture is compared with the results from a system based on Self-Organizing Maps, showing that the growing grid architecture performs significantly superior on the action recognition tasks.     

The Representation of Knowledge in Minds and Machines

Human knowledge can be represented as a propositional network in which the meaning of a node is defined by its position in the network. That is, the relationship between a node and its neighbours determines how this node is used in language understanding and production, i.e., its meaning. The propositions that make up such a network are predicate-argument structures with time and location slots. Schemas, frames, and production rules can be expressed in the same formalism. Implications for this contextual view of meaning are discussed. Since the construction of such a propositional network depends on hand coding and is therefore impractical, an alternative automatic statistical procedure is explored that yields a high-dimensional semantic space. Vectors in this space correspond to nodes in the propositional network, in that the meaning of a vector in the Latent Semantic Analysis space is given by its neighbouring vectors in that space.     

Brain network interactions in auditory, visual and linguistic processing

In the paper, we discuss the importance of network interactions between brain regions in mediating performance of sensorimotor and cognitive tasks, including those associated with language processing. Functional neuroimaging, especially PET and fMRI, provide data that are obtained essentially simultaneously from much of the brain, and thus are ideal for enabling one to assess interregional functional interactions. Two ways to use these types of data to assess network interactions are presented. First, using PET, we demonstrate that anterior and posterior perisylvian language areas have stronger functional connectivity during spontaneous narrative production than during other less linguistically demanding production tasks. Second, we show how one can use large-scale neural network modeling to relate neural activity to the hemodynamically-based data generated by fMRI and PET. We review two versions of a model of object processing - one for visual and one for auditory objects. The regions comprising the models include primary and secondary sensory cortex, association cortex in the temporal lobe, and prefrontal cortex. Each model incorporates specific assumptions about how neurons in each of these areas function, and how neurons in the different areas are interconnected with each other. Each model is able to perform a delayed match-to-sample task for simple objects (simple shapes for the visual model; tonal contours for the auditory model). We find that the simulated electrical activities in each region are similar to those observed in nonhuman primates performing analogous tasks, and the absolute values of the simulated integrated synaptic activity in each brain region match human fMRI/PET data. Thus, this type of modeling provides a way to understand the neural bases for the sensorimotor and cognitive tasks of interest.     

The organization of day-care environments: "zone" versus "man-to-man" staff assignments

In a large day-care center, measures of group participation were used to compare how much of a child's time is lost from planned activities during the daily transition from lunch, through the bathroom and dressing areas, to the nap area. Participation measures were taken using the "Zone" and "Man-to-Man" staffing procedures, two typical methods for dividing responsibility among teaching staff. In the "Zone" procedure, each teacher was assigned responsibility for a particular area, and for all children who passed through that area. In the "Man-to-Man" procedure, each teacher was assigned responsibility for shepherding a group of designated children through all activity areas during each transition. The Lunch-to-Nap transition using the Zone staffing assignment was accomplished with a smaller decrease in child participation in planned activities than the transition utilizing the Man-to-Man procedure. Thus, other things being equal, it is recommended that the Zone procedure be used in group-care programs with more than one staff member, with each teacher being responsible for specific activity areas, rather than specific children.     

The role of retrieval structures in memorizing music

This article explores the use of structure in the encoding and retrieval of music and its relation to level of skill. Twenty-two pianists, classified into four levels of skill, were asked to learn and memorize an assigned composition by J. S. Bach (different for each level). All practice was recorded on cassette tape. At the end of the learning process, the pianists performed their assigned composition in a recital setting. The performances were subsequently evaluated by three experienced pianists according to a standardized grading system. From the cassette tapes, values for the frequency with which pianists started and stopped their practice on "structural," "difficult," and "other" bars were obtained. Starts and stops on each bar type were compared across three stages of the learning process. The analyses reveal that all pianists, regardless of level, started and stopped their practice increasingly on structural bars and decreasingly on difficult bars across the learning process. Moreover, the data indicate that starts and stops increased on structural bars and decreased on difficult bars systematically with increases in level of skill. These findings are interpreted and discussed so as to elucidate characteristics of the retrieval structures adopted by musicians in their practice and performance and how the formation and use of retrieval structures develop as a function of expertise. Finally, the elicited values for starts on structural, difficult, and other bars are examined and discussed according to how they relate to the pianists' scores on performance quality.     

Selective influence through conditional independence

Let each of several (generally interdependent) random vectors, taken separately, be influenced by a particular set of external factors. Under what kind of the joint dependence of these vectors on the union of these factor sets can one say that each vector is selectively influenced by “its own” factor set? The answer proposed and elaborated in this paper is: One can say this if and only if one can find a factor-independent random vector given whose value the vectors in question are conditionally independent, with their conditional distributions selectively influenced by the corresponding factor sets. Equivalently, the random vectors should be representable as deterministic functions of “their” factor sets and of some mutually independent and factor-independent random variables, some of which may be shared by several of the functions.     

Perspectives of probabilistic inferences: Reinforcement learning and an adaptive network compared

The assumption that people possess a strategy repertoire for inferences has been raised repeatedly. The strategy selection learning theory specifies how people select strategies from this repertoire. The theory assumes that individuals select strategies proportional to their subjective expectations of how well the strategies solve particular problems; such expectations are assumed to be updated by reinforcement learning. The theory is compared with an adaptive network model that assumes people make inferences by integrating information according to a connectionist network. The network's weights are modified by error correction learning. The theories were tested against each other in 2 experimental studies. Study 1 showed that people substantially improved their inferences through feedback, which was appropriately predicted by the strategy selection learning theory. Study 2 examined a dynamic environment in which the strategies' performances changed. In this situation a quick adaptation to the new situation was not observed; rather, individuals got stuck on the strategy they had successfully applied previously. This "inertia effect" was most strongly predicted by the strategy selection learning theory.     

Uncertainty-dependent activity within the ventral striatum predicts task-related changes in response strategy

Recent neuroimaging work has demonstrated that the ventral striatum (VS) encodes confidence in perceptual decisions. However, it remains unclear whether perceptual uncertainty can signal the need to adapt behavior (such as by responding more cautiously) and whether such behavioral changes are related to uncertainty-dependent activity within the VS. Changes in response strategy have previously been observed following errors and are associated with both medial frontal cortex (MFC) and VS, two components of the performance-monitoring network. If uncertainty can elicit changes in response strategy (slowing), then one might hypothesize that these changes rely on the performance-monitoring network. In the present study, we investigated the link between perceptual uncertainty and task-related behavioral adaptations (response slowing and accuracy increases), as well as how such behavioral changes relate to uncertainty-dependent activity within MFC and VS. Our participants performed a two-choice perceptual decision-making task in which perceptual uncertainty was reported on each trial while behavioral and event-related functional magnetic resonance imaging data were collected. Analysis of the behavioral data revealed that uncertain (but correct) responses led to slowing on subsequent trials, a phenomenon that was positively correlated with increased accuracy. Critically, post-uncertainty slowing was negatively correlated with the VS activity elicited by uncertain responses. In agreement with previous reports, increases in MFC activation were observed for uncertain responses, although MFC activity was not correlated with post-uncertainty slowing. These results suggest that perceptual uncertainty can serve as a signal to adapt one’s response strategy and that such behavioral changes are closely tied to the VS, a key node in the performance-monitoring network.     

题库结构对原始题在线属性标定准确性之影响研究

目前已有研究证明可达阵在认知诊断测验编制中起重要作用,但迄今为止并没有引起普遍注意。本文主要讨论当题库缺少某些可达阵对应的项目类,对原始题的属性向量在线标定的准确性的影响。本文对含6个属性的独立型结构进行了模拟试验,结果显示：如果题库不充要,原始题的属性标定准确性受到影响,题库中非可达阵中项目对标定有一定的弥补作用。间接印证了可达阵在认知诊断题库起到非常重要的作用。     

Using the conditional grade-of-membership model to assess judgment accuracy

Consider the case whereJ instruments are used to classify each ofI objects relative toK nominal categories. The conditional grade-of-membership (GoM) model provides a method of estimating the classification probabilities of each instrument (or judge) when the objects being classified consist of both pure types that lie exclusively in one ofK nominal categories, and mixtures that lie in more than one category. Classification probabilities are identifiable whenever the sample of GoM vectors includes pure types from each category. When additional, relatively mild, assumptions are made about judgment accuracy, the identifiable correct classification probabilities are the greatest lower bounds among all solutions that might correspond to the observed multinomial process, even when the unobserved GoM vectors do not include pure types from each category. Estimation using the conditional GoM model is illustrated on a simulated data set. Further simulations show that the estimates of the classification probabilities are relatively accurate, even when the sample contains only a small percentage of approximately pure objects.The authors thank Max A. Woodbury, Kenneth G. Manton and H. Dennis Tolley for their help and four anonymous Psychometrika reviewers (including an associate editor) for their beneficial expository and technical suggestions. This work was supported by the Dean's Fund for Summer Research, Owen Graduate School of Management, Vanderbilt University.     

Constrained multidimensional unfolding of confusion matrices: Goal point and slide vector models

Two constrained multidimensional unfolding models, the goal point and slide vector models, are proposed for analyzing confusion matrices. In both models, the row and column stimuli are expressed as two sets of points in a low-dimensional space, where the difference vector connecting a column point to the corresponding row point indicates the change in the stimulus representation through a cognitive process. The difference vector is constrained by the hypothesis that the trend in the representational change is invariant across stimuli: the goal point model constrains all difference vectors to point toward a single point, and the slide vector model constrains all difference vectors to be parallel to each other. In both models the coordinates of points are estimated by the maximum-likelihood method. Examples illustrate that the two models allow us to examine hypotheses about invariant trends in representational changes and to grasp such trends from the resulting configurations.     

Producing high-dimensional semantic spaces from lexical co-occurrence

A procedure that processes a corpus of text and produces numeric vectors containing information about its meanings for each word is presented. This procedure is applied to a large corpus of natural language text taken from Usenet, and the resulting vectors are examined to determine what information is contained within them. These vectors provide the coordinates in a high-dimensional space in which word relationships can be analyzed. Analyses of both vector similarity and multidimensional scaling demonstrate that there is significant semantic information carried in the vectors. A comparison of vector similarity with human reaction times in a single-word priming experiment is presented. These vectors provide the basis for a representational model of semantic memory, hyperspace analogue to language (HAL).     

The acquisition of vector knowledge and its relation to self-rated direction sense

Self-rated sense of direction is reliably related to people's accuracy when pointing in the direction of unseen landmarks from imagined or actual perspectives. It is proposed that the cognitive substrate of accurate pointing responses is a vector representation, which is defined as an integrated network of displacement vectors. Experiment 1 isolated the body senses and tested displacement vector formation in a path-integration task. Experiment 2 isolated the visual modality and tested displacement vector formation in a virtual-learning task. Both experiments tested whether people reporting a good sense of direction (GSOD) were more likely to compute displacement vectors than people reporting a poor sense of direction (PSOD). The results showed that both GSOD and PSOD people computed displacement vectors in the path-integration task, but not in the virtual-learning task. When interlandmark relations were visually specified, GSOD people made more accurate pointing responses than PSOD people, adding to a growing body of cognitive correlates of self-rated direction sense.     

Individual classification of elementary school children’s physical activity: A time-efficient,group-based approach to reference measurements

The objective measurement of physical activity using accelerometers is becoming increasingly popular. There is little consensus, however, about how to analyze acceleration data. One promising approach is the use of reference measurements in which the subjects conduct specific activities. This makes it possible to identify data patterns that indicate these activities for each subject. The drawback of this approach is its rather high cost, in terms of both time and money. We propose a new approach in which a group of children conduct the reference measurements at the same time. We trained support vector machine models on the accelerometer data of 70 children (ages 8–11 years) to predict their activities during those reference measurements. We correctly classified activities with an accuracy of 96.9 % when fitting the individual models for each subject, and 87.5 % when fitting general models for all subjects. The obtained accuracies were comparable to results reported in previous reference measurement studies, in which each subject was measured individually. They were higher than the accuracies obtained by the traditional approach, which transfers accelerometer data to counts and classifies those on the basis of predefined cut points. We concluded that our approach can yield a valuable contribution, particularly to studies with larger samples.     

A marginal maximum likelihood method for the vector threshold model to analyze dichotomous choice data

Abstract: At least two types of models, the vector model and the unfolding model can be used for the analysis of dichotomous choice data taken from, for example, the pick any/ n method. The previous vector threshold models have a difficulty with estimation of the nuisance parameters such as the individual vectors and thresholds. This paper proposes a new probabilistic vector threshold model, where, unlike the former vector models, the angle that defines an individual vector is a random variable, and where the marginal maximum likelihood estimation method using the expectation-maximization algorithm is adopted to avoid incidental parameters. The paper also attempts to discuss which of the two models is more appropriate to account for dichotomous choice data. Two sets of dichotomous choice data are analyzed by the model.     

Objective assessment of gender roles: Gender Roles Test (GRT-36)

This study was designed to develop a computerized test to assess gender roles. This test is presented as a decision-making task to mask its purpose. Each item displays a picture representing an activity and a brief sentence that describes it. Participants have to choose the most suitable sex to perform each activity: man or woman. The test (Gender Roles Test, GRT-36) consists of 36 items/activities. The program registers both the choices made and their response times (RTs). Responses are considered as stereotyped when the chosen sex fits stereotyped roles and non-stereotyped when the chosen sex does not fit stereotyped roles. Individual means (RTs) were computed for stereotyped and non-stereotyped responses, differentiating between domestic and work spheres. A "D" score, reflecting the strength of association between activities and sex, was calculated for each sphere and sex. The study incorporated 78 participants (69% women and 31% men) ranging from 19 to 59 years old. The results show that: (a) reading speed does not explain the variability in the RTs; (b) RTs show good internal consistency; (c) RTs are shorter for stereotyped than for neutral stimuli; (d) RTs are shorter for stereotyped than for non-stereotyped responses. Intended goals are supported by obtained results. Scores provided by the task facilitate both group and individual detailed analysis of gender role, differentiating the gender role assigned to men from that assigned to women, at the domestic and work spheres. Obtained data fall within the scope of the genderology and their implications are discussed.