Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 8
Encontrada(s) 1 página(s)

DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Correlations of network trajectories

  • Lacasa, Lucas
  • Rodríguez-García, Jorge Pablo
  • Eguíluz, Víctor M.
Temporal networks model how the interaction between elements in a complex system evolves over time. Just as complex systems display collective dynamics, here we interpret temporal networks as trajectories performing a collective motion in graph space, following a latent graph dynamical system. Under this paradigm, we propose a way to measure how the network pulsates and collectively fluctuates over time and space. To this aim, we extend the notion of linear correlation functions to the case of sequences of network snapshots, i.e., a network trajectory. We construct stochastic and deterministic graph dynamical systems and show that the emergent collective correlations are well captured by simple measures, and we illustrate how these patterns are revealed in empirical networks arising in different domains., .L. and V.M.E. acknowledge funding from projects DYNDEEP (EUR2021-122007), MISLAND (PID2020-114324GB-C22), and Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R&D (MDM-2017-0711) all funded by MCIN/AEI/10.13039/501100011033. J.P.R. is supported by the Juan de la Cierva Formación program (Reference No. FJC2019-040622-I) funded by the Spanish Ministry of Science and Innovation.
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Haros graphs: an exotic representation of real numbers

  • Calero-Sanz, Jorge
  • Luque, Bartolo
  • Lacasa, Lucas
This paper introduces Haros graphs, a construction which provides a graph-theoretical representation of real numbers in the unit interval reached via paths in the Farey binary tree. We show how the topological structure of Haros graphs yields a natural classification of the reals numbers into a hierarchy of families. To unveil such classification, we introduce an entropic functional on these graphs and show that it can be expressed, thanks to its fractal nature, in terms of a generalised de Rham curve. We show that this entropy reaches a global maximum at the reciprocal of the Golden number and otherwise displays a rich hierarchy of local maxima and minima that relate to specific families of irrationals (noble numbers) and rationals, overall providing an exotic classification and representation of the reals numbers according to entropic principles. We close the paper with a number of conjectures and outline a research programme on Haros graphs., J.C. and B.L. acknowledge funding from Spanish Ministry of Science and Innovation under project M2505 (PID2020-113737GB-I00). L.L. acknowledges funding from projects DYNDEEP (EUR2021-122007), MISLAND (PID2020-114324GB-C22) and through the Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R&D (MDM-2017-0711), all of them funded by the Spanish Ministry of Science and Innovation via AEI., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Chaotic renormalization group flow and entropy gradients over Haros graphs

  • Calero-Sanz, Jorge
  • Luque, Bartolo
  • Lacasa, Lucas
Haros graphs have been recently introduced as a set of graphs bijectively related to real numbers in the unit interval. Here we consider the iterated dynamics of a graph operator R over the set of Haros graphs. This operator was previously defined in the realm of graph-theoretical characterisation of low-dimensional nonlinear dynamics, and has a renormalization group (RG) structure. We find that the dynamics of R over Haros graphs is complex and includes unstable periodic orbits or arbitrary period and non-mixing aperiodic orbits, overall portraiting a chaotic RG flow. We identify a single RG stable fixed point whose basin of attraction is the set of rational numbers, associate periodic RG orbits with (pure) quadratic irrationals and aperiodic RG orbits with (non-mixing) families of non-quadratic algebraic irrationals and trascendental numbers. Finally, we show that the entropy gradients inside periodic RG orbits are constant. We discuss the possible physical interpretation of such chaotic RG flow and speculate on the entropy-constant periodic orbits as a possible confirmation of a (quantum field-theoretic) c-theorem applied inside the invariant set of a RG flow., JC and BL acknowledge funding from Spanish Ministry of Science and Innovation under project M2505 (PID2020-113737GB-I00). LL acknowledges funding from project DYNDEEP (EUR2021-122007) and MISLAND (PID2020-114324GB-C22), both projects funded by the Spanish Ministry of Science and Innovation, and Severo Ochoa and Mar´ıa de Maeztu Program for Centers and Units of Excellence in R&D (MDM-2017-0711) funded by MCIN/AEI/10.13039/501100011033., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Lyapunov Exponents for Temporal Networks

  • Caligiuri, Annalisa
  • Eguíluz, Víctor M.
  • Di Gaetano, Leonardo
  • Galla, Tobias
  • Lacasa, Lucas
By interpreting a temporal network as a trajectory of a latent graph dynamical system, we introduce the concept of dynamical instability of a temporal network, and construct a measure to estimate the network Maximum Lyapunov Exponent (nMLE) of a temporal network trajectory. Extending conventional algorithmic methods from nonlinear time-series analysis to networks, we show how to quantify sensitive dependence on initial conditions, and estimate the nMLE directly from a single network trajectory. We validate our method for a range of synthetic generative network models displaying low and high dimensional chaos, and finally discuss potential applications., AC acknowledges funding by the Maria de Maeztu Programme (MDM-2017-0711) and the AEI under the FPI programme. LL acknowledges funding from project DYNDEEP (EUR2021-122007), and LL and VME acknowledge funding from project MISLAND (PID2020-114324GB-C22), both projects funded by the Spanish Ministry of Science and Innovation. This work has been partially supported by the Mar´ıa de Maeztu project CEX2021-001164-M. funded by MCIN/AEI/10.13039/501100011033., With funding from the Spanish government through the "Severo Ochoa Centre of Excellence" accreditation (CEX2021-001164-M)., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Higher-order correlations reveal complex memory in temporal hypergraphs

  • Gallo, Luca
  • Lacasa, Lucas
  • Latora, Vito
  • Battiston, Federico
Many real-world complex systems are characterized by interactions in groups that change in time. Current temporal network approaches, however, are unable to describe group dynamics, as they are based on pairwise interactions only. Here, we use time-varying hypergraphs to describe such systems, and we introduce a framework based on higher-order correlations to characterize their temporal organization. We analyze various social systems, finding that groups of different sizes have typical patterns of long-range temporal correlations. Moreover, our method reveals the presence of non-trivial temporal interdependencies between different group sizes. We introduce a model of temporal hypergraphs with non-Markovian group interactions, which reveals complex memory as a fundamental mechanism underlying the pattern in the data., L.G. and F.B. acknowledge support from the Air Force Office of Scientific Research under award number FA8655-22-1-7025. L.L. acknowledges funding from the Spanish Research Agency (AEI) via projects DYNDEEP (EUR2021-122007), MISLAND (PID2020-114324GB-C22), and the Mar´ıa de Maeztu project CEX2021-001164-M., With funding from the Spanish government through the "Severo Ochoa Centre of Excellence" accreditation (CEX2021-001164-M)., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

An effective theory of collective deep learning

  • Longhi, Stefano
Unraveling the emergence of collective learning in systems of coupled artificial neural networks points to broader implications for machine learning, neuroscience, and society. Here we introduce a minimal model that condenses several recent decentralized algorithms by considering a competition between two terms: the local learning dynamics in the parameters of each neural network unit, and a diffusive coupling among units that tends to homogenize the parameters of the ensemble. We derive an effective theory for linear networks to show that the coarse-grained behavior of our system is equivalent to a deformed Ginzburg-Landau model with quenched disorder. This framework predicts depth-dependent disorder-order-disorder phase transitions in the parameters’ solutions that reveal a depth-delayed onset of a collective learning phase and a low-rank microscopic learning path. We validate the theory in coupled ensembles of realistic neural networks trained on the MNIST dataset under privacy constraints. Interestingly, experiments confirm that individual networks –trained on private data– can fully generalize to unseen data classes when the collective learning phase emerges. Our work establishes the physics of collective learning and contributes to the mechanistic interpretability of deep learning in decentralized settings., We acknowledge funding from project DYN-DEEP (EUR2021-122007) from the Agencia Estatal de Investigación MCIN/AEI/10.13039/501100011033. LL additionally acknowledges funding from project MISLAND (PID2020-114324GB-C22), and María de Maeztu project CEX2021-001164-M., With funding from the Spanish government through the "Severo Ochoa Centre of Excellence" accreditation (CEX2021-001164-M)., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Dynamical stability and chaos in artificial neural network trajectories along training

  • Danovski, Kaloyan
  • Soriano, Miguel C.
  • Lacasa, Lucas
29 pages, 18 figures, The process of training an artificial neural network involves iteratively adapting its parameters so as to minimize the error of the network's prediction, when confronted with a learning task. This iterative change can be naturally interpreted as a trajectory in network space -- a time series of networks -- and thus the training algorithm (e.g. gradient descent optimization of a suitable loss function) can be interpreted as a dynamical system in graph space. In order to illustrate this interpretation, here we study the dynamical properties of this process by analyzing through this lens the network trajectories of a shallow neural network, and its evolution through learning a simple classification task. We systematically consider different ranges of the learning rate and explore both the dynamical and orbital stability of the resulting network trajectories, finding hints of regular and chaotic behavior depending on the learning rate regime. Our findings are put in contrast to common wisdom on convergence properties of neural networks and dynamical systems theory. This work also contributes to the cross-fertilization of ideas between dynamical systems theory, network theory and machine learning., We acknowledge funding from the Spanish Research Agency MCIN/AEI/10.13039/501100011033 via projects DYNDEEP (EUR2021-122007), MISLAND (PID2020-114324GB-C22), INFOLANET (PID2022-139409NB-I00) and the María de Maeztu project CEX2021-001164-M., With funding from the Spanish government through the "Severo Ochoa Centre of Excellence" accreditation (CEX2021-001164-M)., No
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Otros (Other)

Understanding following patterns among high-performance athletes

  • Rodríguez-García, Jorge Pablo
  • Arola-Fernández, Lluís
Professional sports enhance interaction among athletes through training groups, sponsored events and competitions. Among these, the Olympic Games represent the largest competition with a global impact, providing the participants with a unique opportunity for interaction. We studied the following patterns among highly successful athletes to understand the structure of their interactions. We used the list of Olympic medallists in the Tokyo 2020 Games to extract their follower-followee network in Twitter, finding 7,326 connections among 964 athletes. The network displayed frequent connections to similar peers in terms of their features including sex, country and sport. We quantified the influence of these features in the followees choice through a gravity approach capturing the number of connections between homogeneous groups. Our research remarks the importance of datasets built from public exposure of professional athletes, serving as a proxy to investigate interesting aspects of many complex socio-cultural systems at different scales., With funding from the Spanish government through the "Severo Ochoa Centre of Excellence" accreditation (CEX2021-001164-M)., No