Partenaires et organisations internationales
(Anglais)
|
AT, BE, BG, CH, CZ, DE, DK, ES, FI, FR, GR, HU, IE, IL, IT, LT, NL, NO, PL, PT, RO, RS, SI, UK
|
Résumé des résultats (Abstract)
(Anglais)
|
1 Abstract ' Network structure, robustness and adaptivity of organizations', Frank Schweizer, ETH Zürich, Keywords: Adaptivity, robustness, scaling-laws, network structure, heterogeneous systems The main aim of this project was to investigate how the internal structure of a system determines its robustness and adaptivity. This question is of special interest in social, economic and natural systems composed by many interacting units. On the one hand, robustness is the trait of some systems to retain its properties even under external perturbations. On the other, adaptivity is the capability of a system to modify itself to better respond to a changing environment. During this project, we followed two different insights: First, we studied it at the level of the network structure defining the topology of the system. Second, investigated how robustness and adaptivity are influenced by internal properties of the interacting individuals. We studied the first part, by means of two complementary views: (i) by modeling the evolution of informal organizations aiming at recovering stylized facts of these networks. (ii) through the analysis of hierarchical structures that appear in natural and artificial systems which display both, adaptivity and robustness. The first line comprises a microscopic model in the context of informal organizations. In this context, it was shown that individuals would tend to increase their centrality in order to maximize the profits. Thus, and depending upon the external (environmental) conditions, the individuals create or remove links such that allow them to remain central in the network structure. We introduced a minimalistic model that reproduces the structural properties known for these networks. We have shown that in a rapidly changing environment, flat structures are the most adaptive ones; while under constant conditions, well- defined hierarchies are the most efficient. For intermediate values, hierarchical, complex networks are the most efficient ones. We demonstrated that complex hierarchies are efficient structures for some situations. Due to the interdisciplinary insight of this project, for the second line, we studied the topology of hierarchical structures that appear in natural and artificial systems, displaying both adaptivity and robustness. In natural systems, perhaps the most paradigmatic example of a system showing adaptivity and robustness is that of the biological diversity in our planet. We shown that the tree of life is not the result of a random process, but that successful lineages tend to have descendants which continue to be prolific. The biodiversity which has been produced in this way is smaller than the one that would be obtained under random evolution. In artificial systems, many structures are ordered in hierarchies: in computer systems, this ordering is pervasive: from information storage to problem abstraction in software development. As a first step in this line we studied the directory structures in which users sort their information. We studied how to characterize them, improving on previous research on this line. The second approach to the question about robustness and adaptivity, regards on how the microscopic properties of the individuals modify those properties at the global level. A common trait of social systems, is that they are not identical, but heterogeneous. We investigated the role of diversty in social systems in a changing environment. We found that (up to a certain degree) diversity can enhance the adaptivity to the external influence. Heterogeneity can (also) emerge as the system evolves. In particular, we studied how fast a system can reach a consensus state, in which all the individuals share the same state. Consensus is related with the robustness of the system, given that (if starting from a random configuration) the system gets frozen into an ordered state fast, it implies that its initial properties also disappear. We investigated a paradigmatic example of opinion dynamics: the Voter model. In particular, we studied the role of inertia or aging in such a system: the more a unit has been keeping its current state, the less likely it will modify it. Counter-intuitively, we found that inertia (building up heterogeneity in the system) can decrease the time it takes the system to reach consensus. 2 Abstract 'Failure risk propagation in economic and supply networks', Hans-Jakob Lüthi, Frank Schweizer, ETH Zürich, Keywords: multiplicative stochastic processes, fragility, trend reinforcing, failure propagation The aim of this project is to study failure propagation and systemic risk in economic networks. An economic system is seen as a network of firms coupled through dependencies (e.g. credit or supply). We are taking a networks (or ensemble) perspective and describe firms by only two parameters, namely the wealth and the fragility. The individual dynamics for wealth and fragility of a firm is thought to be of a stochastic nature. Size and fragility dynamics are coupled between firms through a network of dependencies. Coupling happens in two ways: (i) via linear dependencies which correlate the stochastic evolutions of connected firms, that means, a firm’s new wealth or fragility is a linear combination of the former values of its neighbors. And (ii) via damage transfer after failure. A firm fails if its fragility hits a threshold, then it is liquidated which might cause some stress to connected firms either by some debt not being payed or by a missing supplier. We found interesting results on the ensemble level for coupled fragility dynamics under trend persistence, and for coupled growth dynamics when growth is multiplicatively stochastic. In both cases we foud that there exists an optimal intermediate level of coupling. For fragility dynamics it reduces the default probability best, and for growth dynamics it maximizes the over all growth rate. For cascading processes of damage transfer we reviewed several cascading processes and reformulated them under a common viewpoint. This enables a combination with other dynamic processes in the future. 3 Abstract 'Robust fitting technique in financial and non-financial time series', Wolfgang Breymann, Zurich University of Applied Sciences Winterthur, Keywords: generalized hyperbolic distributions, risk quantification, extreme events, scaling, fractals Normal mean-variance mixture distributions have received a lot of attention in the financial-modelling and risk-management literature. A very general class is formed by the generalized hyperbolic (GH) family, which can describe frequently observed statistical facts as asymmetry, heavy-tails and non-linear dependence. The main progress during the last year was the development of a method for fitting GH distributions to a set of short time series and the successful application of this procedure to the computation of risk measures for short fund-of-hedge-fund return series. In addition, the development and support of the R-package has been pursued, and return distributions of daily and intra-day commodity futures data have been investigated. 4 Abstract 'Collaborative information filtering for the Internet', Fribourg team, University of Fribourg, Keywords: recommender systems, collaborative filtering, data mining, risk, information Thanks to the Internet and the World Wide Web, we live in a world of many possibilities—we can choose from thousands of movies, millions of books, and billions of web pages. Created to handle this ever increasing amount of information, recommender systems represent a prominent challenge of information science. In our work we studied extensive datasets containing ratings of real users, proposed two novel recommendation methods based on a diffusion-like process on a network, and further improved the original methods to optimize their performance. Here we overview our main achievements. 5 Abstract 'Impact of severe environment changes on population dynamics', Michel Droz, University of Geneva, Keywords: population dynamics, environmental changes, resilience, cooperative phenomena The main goal of our project was to study several forms of risk encountered by populations due to different causes like changing environment, catastrophes or lack of cooperation. Three main problems were approached: i) the dynamics of a single-species population evolving on a two-dimensional space and submitted to different changing environment; ii) how to support cooperation in the context of the Prisonner’s Dilemma by moving influential players; iii) the extinction risk for a dynamical food web model. New results were obtained for all the tree topics, leading to publications in main international research journals.
|
