View: session overviewtalk overview

12:00 | Thermodynamic efficiency of autocatalytic networks ABSTRACT. An autocatalytic network is a reaction network modelling a set of chem- ical species that mutually catalyse each other’s production through chem- ical reactions, starting from a finite set of species assumed to be available from the environment. Autocatalytic networks are capable of spon- taneous emergence and self-reproduction, and are supposed to be chemical networks at the basis of life. In this work, I investigate the thermodynamic features of autocatalytic networks. In particular, I use recent results on nonequilibrium thermody- namics of Chemical Reaction Networks in order to study the con- nection between the topological constrains a network must satisfy to be autocatalytic, and its ability to store the chemical work performed by the surrounding environment, i.e., its thermodynamic efficiency: eta = 1−EP/Wc. Here EP is the entropy production and Wc is the chemical work performed by the environment on the chemical network. By studying the evolution of simple autocatalytic networks, I observe that the topology of an autocatalytic network does not strongly constrain the thermodynamic properties of the system, allowing different networks to exhibit various thermodynamic behaviours. Furthermore, the obtained results suggest that the effects of catalyses on thermodynamic flows may result in an advantage for autocatalytic networks over other systems in storing energy. |

12:15 | Helios-Web: An interactive framework to explore large complex networks on the web ABSTRACT. Network science emerged as a proper framework to represent, analyze and model a diverse range of complex systems. Still, investigating such structures can become considerably challenging when they encompass many nodes and edges. Moreover, it is typical these elements to be associated with metadata and other features, further increasing the complexity and dimensionality of the problem. In this context, interactive visualization can streamline the data exploration process by facilitating the discovery of hidden patterns and providing intuitive interpretations of the system under analysis. However, most of the existing tools can not handle large networks comprising more than 10,000 nodes in real-time, which is a reflection of their lack of both rendering capabilities and continuous layout algorithms. Here, we developed a new network visualization and exploration tool, Helios-web ( https://github.com/filipinascimento/helios-web ) that incorporates GPU-based rendering and continuous force-directed layouts that can visualize networks of millions of nodes. This is attained by using a variety of rendering techniques, such as billboards, signed distance fields, and GPU-based picking. The tool also includes an API and interactivity features to search, filter and highlight nodes or edges according to their associated attributes. Helios-web is being developed for a web environment and can be integrated into portals and websites. We illustrate the usefulness of the tool through an exploration of scholarly networks obtained from the APS, Microsoft Academic Graph, and OpenAlex datasets. |

12:30 | Phishing Detection on Ethereum via Attributed Ego-graph Embedding ABSTRACT. In recent years, the losses caused by phishing scams on Ethereum have reached a level that cannot be ignored. In such a phishing detection scenario, network embedding is seen as an effective solution. In this paper, we propose an attributed ego-graph embedding framework to distinguish phishing accounts. We first obtain the account labels from an authority site and the transaction records from Ethereum on-chain blocks. Then we extract ego-graphs for each labeled account to represent it. To learn representations for ego-graphs, we utilize non-linear substructures sampled from ego-graphs and use a skip-gram model. Finally, a classifier is applied to graph embeddings to predict phishing accounts. To overcome the limit that transaction attributes are not encoded into ego-graph embeddings, we give nodes and subgraphs with richer attribute-based semantics. Specifically, we propose a novel node relabeling strategy based on Ethereum transaction attributes including transaction amount, number, and direction, and differentiating nodes and subgraphs by new labels. Through this, structural and attributed features of the Ethereum transaction networks can be learned at the same time. Experimental results show that our framework achieves effective performance on class imbalanced phishing detection on Ethereum. |

12:45 | A multiplex analysis of phonological and orthographic networks ABSTRACT. The study of natural language using a network approach has made it possible to characterize novel properties ranging from the level of individual words to phrases or sentences. A natural way to quantitatively evaluate similarities and differences between spoken and written language is by means of a multiplex network defined in terms of a similarity distance between words. Here, we use a multiplex representation of words based on orthographic or phonological similarity to evaluate their structure. As a continuation of our work [Lara-Martínez, P., Obregón-Quintana, B., Reyes-Manzano, C. F., López-Rodríguez, I., & Guzmán-Vargas, L. (2021). Comparing phonological and orthographic networks: A multiplex analysis. Plos one, 16(2), e0245263.], we adjust the way of assigning links so that these links show the morphological similarity in a better way. We find that from the analysis of topological properties of networks, there are different levels of local and global similarity when comparing written vs. spoken structure across 12 natural languages from 4 language families (Romance, Germanic, Slavic and Uralic), by analyzing 50000 different words of each language. In particular, it is found that differences between the phonetic and written layers is markedly higher for French and English, while for the other languages analyzed, this separation is relatively smaller. Specifying some results, we find the following: (i) a Weibull behavior in the degree distribution, (ii) clustering by degree distribution retrieves the language family, (iii) the behavior of each language in finding communities that maximize modularity is similar in each language family separately, and (iv) differences between linguistic families in robustness (for example a marked difference in the Germanic language between layers). We conclude that the multiplex approach allows us to explore additional properties of the interaction between spoken and written language. |

13:00 | Peer learner networks impact study-abroad second language acquisition: Insights from mixed-methods SNA PRESENTER: Andrzej Jarynowski ABSTRACT. Social networks play a vital role in SLA. Combining computational and anthropological Social Network Analysis (SNA), we investigate the influence of peer interaction dynamics and social graph topology on measurable outcomes in two intensive language courses: a 5-week course of German for Erasmus+ exchange students in Baden-Württemberg (n=40), and two editions of a 4-week summer course of the Polish language and culture in Warsaw (n=332). Unlike previous Study Abroad social network research concentrating on i) the micro-level of individual learners’ egocentric networks, presenting an emic view only, and ii) primarily TL native-speaker contacts, we demonstrate how and why peer learner networks can be examined in their entirety, complementing an etic perspective. In particular, we focus on the moderating role of the social network (mesoscopic explanatory variable)—in turn influenced by engagement with the TL-speaking culture (macroscopic explanatory variable)—on L2 progress (microscopic response variable). The study addresses the following overarching questions: RQ1: Is the learners’ L2 progress influenced by their position in the peer interaction network (center vs. periphery) and community membership? RQ2: Which types of interaction revealed in the social graph structure are the most important predictors of L2 progress: - unidirectional or reciprocal? - overall (irrespective of the language(s) used) or in the TL? - incoming or outgoing? RQ3: With respect to TL use, is a more important factor the absolute numbers of immersion hours in the language, or the proportion of L2 use to total communication? RQ4: Is there a relationship between participants’ language progress and the intensity of their contacts with same-L1 users (investigation of homophily effects; cf. Lazarsfeld & Merton, 1954; McPherson et al., 2001)? RQ5: Do the students prefer to socialize with peers demonstrating a similar or different level of L2 proficiency? RQ6: Is TL progress conditioned by network-external factors such as motivation or competence in other (background) languages? The quantitative component of the project showed among others i) that outgoing interactions in the TL are a stronger predictor of progress than incoming interactions, ii) a clear detrimental effect of interactions with same-L1 speakers (routgoing=−.31[-0.63, 0.00],p=.048), iii) the strongest influence of the network in the domains of pronunciation and lexis, where degree centrality in the TL positively correlates with progress (routdegree=.258,p=.001 for pronunciation; routdegree=.304,p=.0002 and rindegree=.263,p=.001 for vocabulary), while betweenness in total communication is significantly anticorrelated (r=−.242,p=.003 and r=−.204,p=.01, respectively). iv) This mirrors the impact of closeness centrality (ease of access to other students). v) Combined with the deleterious influence on SLA of a high in-degree, this underscores the importance of the network’s structural properties. In turn, structured interviews carried out with course participants and their instructors yielded valuable information on the formation and types of the networks the learners engaged in and the purposes these networks served. The presentation will thus illustrate the benefits of combining computational (quantitative) and anthropological (qualitative) social network analysis. Lastly, we shall also compare two face-to-face iterations of one of the courses with its online edition during the COVID-19 pandemic. |

13:15 | Variable selection aided by correlation networks: Revealing behavioral immunity landscapes ABSTRACT. Mapping in vivo the different states of immune cells is an important open problem in biology. In this work, using data from movement and shape of over 100\,000 cells, we found that supervised selection of morpho-kinetic variables guided by a combination of wrapper and filter methods captured the behavioral landscapes of active inflammation [1]. Our mathematical modeling, based on logistic and decision tree models, as well as correlation networks, provided distributions on graphs which allowed us to identify the most important variables for immune cell prediction. The Euclidean projection was made using multidimensional scaling, employing Pearson’s distances as entries of the dissimilarity matrix. |

12:00 | Switchover phenomenon induced by epidemic seeding on geometric networks ABSTRACT. It is a fundamental question in disease modeling how the initial seeding of an epidemic, spreading over a network, determines its final outcome. One important goal has been to find the seed configuration, which infects the most individuals. Although the identified optimal configurations give insight into how the initial state affects the outcome of an epidemic, they are unlikely to occur in real life. In this talk, we identify two important seeding scenarios, both motivated by historical data, that reveal a complex phenomenon [1]. In one scenario, the seeds are concentrated on the central nodes of a network, while in the second one, they are spread uniformly in the population. We challenge the intuition that an epidemic from the most tightly connected, central nodes of a network always leads to a larger epidemic in the long run, in terms of the number of final infected people. Through data-driven and synthetic simulations on real and modelled geometric metapopulation networks, we find that when the basic reproduction number R0 is close to 1, more individuals become infected in the central seeding scenario, but for larger values of R0, the uniform seeding scenario is more dangerous (see the figure). We find that this switchover phenomenon is amplified by the geometric nature of the underlying network by simulations on a variant of the Hyperbolic Random Graph model [2], which features both spacial properties and a scale-free degree distribution, similarly to real metapopulation networks. Based on an equivalence between metapopulation models and bond percolation established by [3], we also provide mathematically rigorous proofs of the switchover phenomenon in analytically tractable network models such as the Configuration Model and the Stochastic Block Model. Our results expand on the previous finding that in case of a single seed, the central scenario is always more dangerous, and further our understanding why the sizes of consecutive waves of a pandemic can differ even if their epidemic characters are similar. We highlight the importance to follow not only the rate but also the spatial distribution of new infection cases during an ongoing pandemic. References [1] G.Ódor, D.Czifra, J.Komjáthy, L.Lovász, and M.Karsai,“Switchover phenomenon induced by epidemic seeding on geometric networks,” Proceedings of the National Academy of Sciences, vol. 118, no. 41, 2021. [2] D. Krioukov, F. Papadopoulos, M. Kitsak, A. Vahdat, and M. Boguná, “Hyperbolic geometry of complex networks,” Physical Review E, vol. 82, no. 3, p. 036106, 2010. [3] M. Barthélemy, C. Godreche, and J.-M. Luck, “Fluctuation effects in metapopulation models: percolation and pandemic threshold,” Journal of theoretical biology, vol. 267, no. 4, pp. 554–564, 2010. |

12:15 | Modeling human proximity networks with random hyperbolic graphs PRESENTER: Fragkiskos Papadopoulos ABSTRACT. Proximity networks are time-varying graphs representing the closeness among humans moving in a physical space. Understanding their properties is critical because they affect the behavior of disease and information spreading, as well as the formation and evolution of communities. Interestingly, a simple model of mobile agents, which assumes the agents perform random walks, reproduces a wide variety of these properties. However, until recently the mechanisms responsible for more complex social behavior remained elusive. Specifically, random walks cannot reproduce the recurrent formation of groups of the same people, which originate from human motion patterns that are far from random. We show that many of the properties of proximity networks, including the formation of recurrent groups (components), emerge naturally and simultaneously in a simple latent space model, called dynamic-S1. The dynamic-S1 does not model agent mobility, but captures their connectivity in each snapshot--each snapshot in the model is a realization of the S1 model of traditional complex networks, which is isomorphic to random hyperbolic graphs. By forgoing the motion component the model facilitates mathematical analysis, allowing us to prove the contact, inter-contact and weight distributions. We show that these distributions are power laws in the thermodynamic limit with exponents lying within the ranges observed in real systems. Furthermore, the behavior of compartmental epidemic spreading processes, such as SIS and SEIR, is remarkably similar in real and modeled networks. We have also shown that the time-aggregated representation of real human proximity networks can be meaningfully embedded into hyperbolic space, using methods developed for the S1 model. Using the resulting embeddings one can identify communities, facilitate greedy routing on the temporal network, and predict future links. Taken altogether, our results indicate that dynamic random hyperbolic graphs are adequate null models of human proximity networks. |

12:30 | Dimension matters when modelling network communities in hyperbolic spaces ABSTRACT. See attached PDF file. |

12:45 | Nearest-neighbour network in hyperbolic space: a simple model of a directed network with asymmetric degree distribution ABSTRACT. Reference and recommendation networks are ubiquitous: encyclopedia articles and scientific papers refer to each other, people recommend each other books, films and music, online shops are full of “people who like this also like that” recommendations. Such networks are substantially asymmetric: the rules according to which a node becomes a source are different from those according to which it becomes a target. Often, the number of recommendations given is effectively bounded, while the number of times a node is recommended (in-degree) is unlimited and has a wide distribution. Here we suggest a relatively simple model of a directed network with asymmetric degree distribution: narrow distribution of out degree and (generally speaking) wide distribution of the in degree, and study its structural properties. Our model belong to the hyperbolic class [1,2] and is a direct generalization of nearest-neighbor models studied extensively for the case of Euclidean metric spaces. Consider a disk on a hyperbolic space, put a large number of points onto the disk uniformly at random, and then connect each point by directed links to a fixed number $m$ of its nearest neighbors. We study the properties of resulting networks in the limit when the area of the disk diverges while the dimensionless density of the points (measured in the space curvature units) remains constant. We show that the resulting networks consist of two distinct parts (see figure), a central core where the in-degree has an approximately Poissonian distribution with average $m$, and peripheral part, where the average in-degree is position-dependent and increases exponentially with increasing distance from the periphery of the disk. The distribution of nodes between core and periphery is controlled by the dimensionless density of the nodes: in high-density networks the core dominates, while the low-density are dominated by the periphery. The resulting overall in-degree distribution is a truncated power law with exponent -3, the width of the power-law region depends on the density, in the limit of low density the distribution becomes a true power law. Notably, in contrast with the model presented in [1] average in-degree of the network remain finite and equal to $m$ in the thermodynamic limit. Similarly to [1] the exponent of the in-degree distribution can be regulated by switching to a quasi-uniform distribution of nodes. We calculate additional structural properties of the network, such as the fraction of bidirectional links, and show how it can be regulated by introducing an additional temperature-like parameter governing the probability of link formation. 1. Krioukov, D., Papadopoulos, F., Kitsak, M., Vahdat, A., Boguñá, M., Hyperbolic geometry of complex networks, Phys. Rev. E 82, 036106 (2010) . 2. Papadopoulos, F., Kitsak, M., Serrano, M.Á., Boguná, M., Krioukov, M. Popularity versus similarity in growing networks, Nature 489, 537 (2012). |

13:00 | Model-independent embedding of directed networks into geometric spaces PRESENTER: Bianka Kovács ABSTRACT. Hyperbolic network models are known to be able of generating networks that are scale-free, highly clustered, and small-world, and can even have a pronounced community structure as well. The success of the hyperbolic network models provides a strong motivation for the development of hyperbolic embedding algorithms that tackle the problem of finding the optimal hyperbolic coordinates of the nodes based on the network topology. However, the previous hyperbolic embedding methods were developed for undirected graphs and can not grasp the asymmetry of the connections occurring in directed networks. Here, we propose a general framework for embedding directed networks into both Euclidean and hyperbolic spaces, where two positions are assigned for each node in the system: one representing its behaviour as a source and one that characterises it as a target of links. Contrary to previous methods, our newly introduced Euclidean-hyperbolic conversion transforms Euclidean embeddings into hyperbolic ones without assuming any specific hyperbolic model as the generator of the network to be embedded. |

13:15 | Scaling up real networks by geometric branching growth ABSTRACT. Real networks often grow through the sequential addition of new nodes that connect to older ones in the graph. However, many real systems evolve through the branching of fundamental units, whether those be scientific fields, countries, or species. Here, we provide empirical evidence for self-similar growth of network structure in the evolution of real systems---the journal citation network and the world trade web---and present the Geometric Branching Growth model, which predicts this evolution and explains the symmetries observed. The model produces multiscale unfolding of a network in a sequence of scaled-up replicas preserving network features, including clustering and community structure, at all scales. Practical applications in real instances include the tuning of network size for best response to external influence and finite-size scaling to assess critical behavior under random link failures. |

13:45 | Complex Contagion: Unfolding and Control |

14:20 | How do ecological networks respond to global change? |

Yamir Moreno & Raissa D'Souza