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.     

Priming effects of arithmetic signs in 10- to 15-year-old children

In this research, 10- to 12- and 13- to 15-year-old children were presented with very simple addition and multiplication problems involving operands from 1 to 4. Critically, the arithmetic sign was presented before the operands in half of the trials, whereas it was presented at the same time as the operands in the other half. Our results indicate that presenting the ‘x’ sign before the operands of a multiplication problem does not speed up the solving process, irrespective of the age of children. In contrast, presenting the ‘+’ sign before the operands of an addition problem facilitates the solving process, but only in 13 to 15-year-old children. Such priming effects of the arithmetic sign have been previously interpreted as the result of a pre-activation of an automated counting procedure, which can be applied as soon as the operands are presented. Therefore, our results echo previous conclusions of the literature that simple additions but not multiplications can be solved by fast counting procedures. More importantly, we show here that these procedures are possibly convoked automatically by children after the age of 13 years. At a more theoretical level, our results do not support the theory that simple additions are solved through retrieval of the answers from long-term memory by experts. Rather, the development of expertise for mental addition would consist in an acceleration of procedures until automatization.     

A Social Interpolation Model of Group Problem-Solving

How do people use information from others to solve complex problems? Prior work has addressed this question by placing people in social learning situations where the problems they were asked to solve required varying degrees of exploration. This past work uncovered important interactions between groups' connectivity and the problem's complexity: the advantage of less connected networks over more connected networks increased as exploration was increasingly required for optimally solving the problem at hand. We propose the Social Interpolation Model (SIM), an agent-based model to explore the cognitive mechanisms that can underlie exploratory behavior in groups. Through results from simulation experiments, we conclude that “exploration” may not be a single cognitive property, but rather the emergent result of three distinct behavioral and cognitive mechanisms, namely, (a) breadth of generalization, (b) quality of prior expectation, and (c) relative valuation of self-obtained information. We formalize these mechanisms in the SIM, and explore their effects on group dynamics and success at solving different kinds of problems. Our main finding is that broad generalization and high quality of prior expectation facilitate successful search in environments where exploration is important, and hinder successful search in environments where exploitation alone is sufficient.     

On the Generalization of Simple Alternating Category Structures

A fundamental question in the study of human cognition is how people learn to predict the category membership of an example from its properties. Leading approaches account for a wide range of data in terms of comparison to stored examples, abstractions capturing statistical regularities, or logical rules. Across three experiments, participants learned a category structure in a low-dimension, continuous-valued space consisting of regularly alternating regions of class membership (A B A B). The dependent measure was generalization performance for novel items outside the range of the training space. Human learners often extended the alternation pattern––a finding of critical interest given that leading theories of categorization based on similarity or dimensional rules fail to predict this behavior. In addition, we provide novel theoretical interpretations of the observed phenomenon.     

The role of variety in the evolution of information society

Information is power. There are several theoretical considerations which guide the basic research on information society and its evolution. Some of them, worked out in this paper, deal with the role of social variety in the process of post‐industrial evolution. Exploring similarities in bio‐genetic information, the process of social transformation is approached by using social system theory.

With certain analogies to biological transformations—mutations—important elements for the transformation of information‐guided behaviour systems (cultures) may be observed. They create a variety of elements (sub‐systems), which respond for changes in the social layout of all kinds of organisations like states, partys, groups, insofar as these changes aim the optimation of functional effectivity in dependence of the variety and its use by a social system's elements (events, acts).

By information processing, collectivities are able to explote even subtile fluctuations in socio‐cultural information variety to “cause” changes. Collectivities are able to instruct and functionalise emerging information through feedback in such a way, that they get better chances to reach certain local values of selection. They are fitter in the sense of better instructed. This process is self‐organised, based on the principle of selection, so that the system chooses out of a complex landscape of values trough communication of measuring parameters like “good” and “bad”. The value‐system is learned through socialisation and changed by social experience.

If social change and variety are optimised, the replication of the system works close to a “treshold of error”, that means that transformation processes are driven to a dynamic equilibrium between stabiliy and creativity. A variety of new transformation operators emerges. Operators of the former dominant information code are still reproducted, but a lot of other alternatives, stabilised by mutual separation are also represented in a survivable concentration and quantity. The competition between these new operators can be transformed in cooperation by their cyclic boundage in networks.

Concluding the paper argues that social systems which intend an optimation of their power management functions by integration of information variety, take advantage in their development in comparison to those who neglect this aspect. They are prepared to develop in a changing environment. Their optimation is a result of feedback within a variety of leadership‐systems, which may also be seen as competitive lifestyles. At a certain critic point there occurs a break in the symmetry of the dominant organisation: a multipole political leadership system emerges, behaviours are derepressed and activated. Uniformity in social time is broken and with it the linear thinking.     

Modelling the effects of semantic ambiguity in word recognition

Most words in English are ambiguous between different interpretations; words can mean different things in different contexts. We investigate the implications of different types of semantic ambiguity for connectionist models of word recognition. We present a model in which there is competition to activate distributed semantic representations. The model performs well on the task of retrieving the different meanings of ambiguous words, and is able to simulate data reported by Rodd, Gaskell, and Marslen-Wilson [J. Mem. Lang. 46 (2002) 245] on how semantic ambiguity affects lexical decision performance. In particular, the network shows a disadvantage for words with multiple unrelated meanings (e.g., bark) that coexists with a benefit for words with multiple related word senses (e.g., twist). The ambiguity disadvantage arises because of interference between the different meanings, while the sense benefit arises because of differences in the structure of the attractor basins formed during learning. Words with few senses develop deep, narrow attractor basins, while words with many senses develop shallow, broad basins. We conclude that the mental representations of word meanings can be modelled as stable states within a high-dimensional semantic space, and that variations in the meanings of words shape the landscape of this space.     

Perception of Happening: How the Brain Deals with the No-History Problem

In physics, the temporal dimension has units of infinitesimally brief duration. Given this, how is it possible to perceive things, such as motion, music, and vibrotactile stimulation, that involve extension across many units of time? To address this problem, it is proposed that there is what is termed an “information construct of happening” (ICOH), a simultaneous representation of recent, temporally differentiated perceptual information on the millisecond time scale. The main features of the ICOH are (i) time marking, semantic labeling of all information in the ICOH with ordinal temporal information and distance from what is informationally identified as the present moment, (ii) vector informational features that specify kind, direction, and rate of change for every feature in a percept, and (iii) connectives, information relating vector informational features at adjacent temporal locations in the ICOH. The ICOH integrates products of perceptual processing with recent historical information in sensory memory on the subsecond time scale. Perceptual information about happening in informational sensory memory is encoded in semantic form that preserves connected semantic trails of vector and timing information. The basic properties of the ICOH must be supported by a general and widespread timing mechanism that generates ordinal and interval timing information and it is suggested that state-dependent networks may suffice for that purpose. Happening, therefore, is perceived at a moment and is constituted by an information structure of connected recent historical information.     

A General Structure for Legal Arguments About Evidence Using Bayesian Networks

A Bayesian network (BN) is a graphical model of uncertainty that is especially well suited to legal arguments. It enables us to visualize and model dependencies between different hypotheses and pieces of evidence and to calculate the revised probability beliefs about all uncertain factors when any piece of new evidence is presented. Although BNs have been widely discussed and recently used in the context of legal arguments, there is no systematic, repeatable method for modeling legal arguments as BNs. Hence, where BNs have been used in the legal context, they are presented as completed pieces of work, with no insights into the reasoning and working that must have gone into their construction. This means the process of building BNs for legal arguments is ad hoc, with little possibility for learning and process improvement. This article directly addresses this problem by describing a method for building useful legal arguments in a consistent and repeatable way. The method complements and extends recent work by Hepler, Dawid, and Leucari (2007) on object‐oriented BNs for complex legal arguments and is based on the recognition that such arguments can be built up from a small number of basic causal structures (referred to as idioms). We present a number of examples that demonstrate the practicality and usefulness of the method.