Abstract: In recent years, the increasing availability of publication records has attracted the attention of the scientific community. In particular, many efforts have been devoted to the study of the organization and evolution of science by exploiting the textual information in the articles like words extracted from the title and abstract. However, just few of them analyzed the main part of an article, i.e., its body. The access to the whole text, instead, allows to pinpoint related papers according to their content, analyzing the network of similarity between them. Scientific concepts are extracted from the body of the articles available in the ScienceWISE platform, but the paper similarity network possesses a considerably high link density (36 %) which spoils any attempt of associating communities of papers to topics. This happens because not all the concepts inside an article are truly informative and, even worse, they may not be useful to discriminate articles with different contents. The presence of such ``generic'' concepts with a loose meaning implies that a considerable amount of connections is made up by spurious similarities. To eliminate generic concepts, we introduce a method to evaluate the concepts' relevance according to an information-theoretic approach. The significance of a concept $c$ is defined in terms of the distance between its maximum entropy, $S_{max}$, and the actual one, $S_c$, calculated using its frequency of occurrence inside papers $tf_c$. Generic concepts are automatically identified as the ones with an entropy close to their maximum and disregarded, while only ``meaningful'' concepts are retained when constructing the paper similarity network. Consequently, the number of links decreases, as well as the amount of noise in the strength of articles' similarities. Hence, the resulting network displays a more well defined community structure, where each community contains articles related to a specific topic.

Abstract: Ants use visual cues for their navigation. There are many studies reporting that ants cannot use map-like systems. Rather, they appear to adopt taxon-systems in which one-to-one correspondences between two different landmarks are realized. To this end, ants can use different paths on their inbound from paths on their outbound. Actually, they appear to head for or ignore visual cues depending on whether they consume food or not. In ant navigation systems therefore, arriving at one location after passing a certain visual cue can be strengthened via several training, resulting in associating each location with a certain visual landmark. After several foraging trips, ants might head for learned visual cues in order to reach goal locations. In this learning mechanism, there is no room for considering other landmarks when foragers head for certain locations. However, several studies reported that ants can exhibit latent learning. This phenomenon can be related to the problem whether or not foragers on inbound trips can use visual cues acquired on their outbound in order to return to their nest. Originally, several trips enhance the relationship between nest location and visual cues. However, in this case, latent learned cues on their outbound can be directly applied to inbound navigation systems. Our aim of this study is checking whether ants can establish effective foraging strategies by associating disconnected information with each other. In this paper, by exposing Japanese wood ants to right-angle-shaped maze or linear-shaped maze on their outbound, we observed trajectories of foragers on their initial inbound trips. On inbound trips, mazes were removed. Thus, foragers could move freely on the test arena. We found that foragers were able to follow their outbound paths when they were restricted to right-angle-shaped maze on their outbound compared with linear-shaped maze on their outbound.

Abstract: This contribution presents a Network-Oriented Modelling approach based on temporal-causal networks. The temporal-causal modelling approach incorporates a dynamic perspective on causal relations. Basic elements are networks of nodes and connections with for each connection a connection weight for the strength of the impact of the connection, for each node a speed factor for the timing of the effect of the impact, and for each node the type of combination function used to aggregate multiple impacts on this node. The approach covers specific types of neural networks, but it is more generic; it also covers, for example, probabilistic and possibilistic approaches in which product or max and min-based functions are used. The temporal-causal network modelling format used enables to address complex phenomena such as the integration of emotions within all kinds of cognitive processes, of internal simulation and mirroring of mental processes of others, and of social interactions. Also adaptive networks are covered in which connection weights of the network change over time, which, for example, can be used to model Hebbian learning in adaptive neuro-cognitive models or evolving social interactions. By choosing suitable combination functions every process that can be modelled as a smooth state-determined system by first-order differential equations, also can be modelled by the presented temporal-causal network modelling approach. At the European Conference on AI ECAI’16, a tutorial is organised about the temporal-causal network modelling approach [1]. Moreover, [2] is a journal paper about the approach, and a book [3] on the approach will be published by Springer in the series Understanding Complex Systems. [1] http://www.few.vu.nl/~treur/ECAI16tutorialTCNM/ [2] Treur, J., Dynamic Modeling Based on a Temporal-Causal Network Modeling Approach. Biologically Inspired Cognitive Architectures, 16, 131-168 (2016) [3] Treur, J., Network-Oriented Modelling: Addressing Complexity of Cognitive, Affective and Social Interactions. Series on Understanding Complex Systems, Springer Publishers, 2016, to appear.

Abstract: Research on the origins of power law and observation and validation of power law distribution in empirical data is active in recent years. This paper is an interdisciplinary research, focusing on exploring the power law distribution in school dropout data for the State of Pennsylvania in the United States. By using the fitting method with goodness-of-fit test based on the Kolmogorov- Smirnov statistics and least-squares fitting, the data is tested in two types of power law distributions—the survival (rank) distribution (“Zipf distribution”) and the complementary cumulative probability distribution (CCDF). In both distributions, only the middle range of the data shows power law and the upper quantile of the distribution bends down from the power law fit. It has two implications: first, technical-wise, it reflects issues of empirical data to fit in the power law since empirical data is affected by the availability of datasets and in social systems, data is bounded by other societal factors. Second, social science wise, it indicates that the dropout rates obey a skewed distribution, which means that the average value of dropout rates loses its meaning. This paper argues that policy makers and researchers should instead focus on the extreme values of dropout rates to better understand the high dropout rates phenomenon in certain districts and areas.

Abstract: Complex systems science studies systems made of many components, where key information about system properties and structures are often conveyed through complex two-dimensional visualization. Such 2D visualization can be very difficult to understand, or even inaccessible at all, for learners whose sensory and/or cognitive modes are not compatible with it, including blind and visually impaired learners and learners who are more successful in understanding abstract materials through physical interaction with concrete, tangible objects. The importance, and difficulty, of making complex visualization more accessible to broader audience has been noted in STEM education literature, but not much development has been made to address this problem yet. Here we explore possibilities of making complexity more accessible to broader participants through three-dimensional manipulatable representations (3D “physicalization”) using 3D printing technologies. We assume that, given the optimal design, materials and 3D printing processes, 3D physicalization will substantially improve the learning of complex systems concepts for a variety of learners, compared to using 2D visualization only. We have conducted iterative designs of the following two physicalizations so far: (1) complex network diagrams, and (2) trajectories of chaotic systems. Several iterations of design and testing have revealed non-trivial design challenges. For example, conventional network embedding algorithms (e.g., spring embedding) are not suitable for physicalization as they tend to embed important nodes in inaccessible areas, which we have resolved by developing heuristic layout algorithms that place all nodes on a hollow (semi-)sphere. Another example of the challenges is how to make crowded parts haptically discernible (e.g., trajectories of strange attractors). This illustrates the importance of striking a right balance between scientific accuracy and pedagogical clarity when we physicalize complex systems. Future work includes further design optimization, experimental evaluation of educational effects of 3D physicalization on learners with diverse backgrounds/abilities, and applications to actual complex systems problem solving.