arXiv subjects

Luc Berthouze

Publications and source records attributed to Luc Berthouze.

At least 19 recordsLinked to original sources

arXiv2026

Tempest: A GPU-Accelerated Engine for Streaming Temporal Random Walks

Temporal random walks, which sample causality-preserving paths, are widely used to analyze time-stamped interactions in domains such as microservices, finance, and online platforms. Generating such walks at scale is challenging because real-world graphs evolve as high-volume streams, making continuous ingestion, efficient memory usage, and strict temporal ordering essential for practical deployment. We present Tempest (TEMPoral nEtwork Streaming Traversals), a GPU-accelerated engine for streaming temporal random walks. Tempest combines a GPU-native dual-index organization over a shared edge store with a hierarchical cooperative scheduler that dispatches walks at thread, warp, or block granularity based on per-step node convergence, enabling efficient start-edge selection, hop-by-hop causality enforcement, and window-based eviction without synchronization. It further provides closed-form constant-time samplers for common temporal bias functions. Our evaluation demonstrates sustained real-time processing of billion-edge streams under sliding windows, outperforming prior systems in ingestion and walk generation throughput while preserving causal correctness.

cs.DC↗

arXiv2026

Heterophily as a generative mechanism for self-organized synergistic interdependencies

Understanding what and how causal dynamical mechanisms generate collective phenomena is a central challenge in complexity science. Recent studies have focused on identifying the mechanisms underlying the synergistic interdependencies that characterise these phenomena in systems with fixed interaction structures. Yet, real-world systems displaying collective phenomena, such as brains, societies, and ecosystems, are adaptive: interactions change in time. Here, we show that heterophily is a minimal local adaptive mechanism for the emergence of self-organized synergistic interdependencies. We study a paradigmatic spin-glass-like model with co-evolving couplings to show how heterophily generates the conditions for synergy to emerge. By solving the minimal $N=3$ case analytically, we reveal the precise mechanism: heterophily weakens pairwise dependencies while inducing high-order dependencies via geometric constraints on the configurations it selects. Together, these two effects underpin synergy. Numerical simulations confirm that this mechanism persists in large systems and that it is robust under parameter heterogeneities and dynamics. We demonstrate the applicability of our results by showing how heterophily can disrupt polarization while promoting synergistic information dynamics of opinions, where individuals' opinions are better explained by group-level influences than by pairwise ones. These results offer a parsimonious route to self-organized synergistic interdependencies in information-processing systems, with potential applications in computational social science, neuroscience, and biology.

physics.soc-ph↗

arXiv2025

FC-ADL: Efficient Microservice Anomaly Detection and Localisation Through Functional Connectivity

Microservices have transformed software architecture through the creation of modular and independent services. However, they introduce operational complexities in service integration and system management that makes swift and accurate anomaly detection and localisation challenging. Despite the complex, dynamic, and interconnected nature of microservice architectures, prior works that investigate metrics for anomaly detection rarely include explicit information about time-varying interdependencies. And whilst prior works on fault localisation typically do incorporate information about dependencies between microservices, they scale poorly to real world large-scale deployments due to their reliance on computationally expensive causal inference. To address these challenges we propose FC-ADL, an end-to-end scalable approach for detecting and localising anomalous changes from microservice metrics based on the neuroscientific concept of functional connectivity. We show that by efficiently characterising time-varying changes in dependencies between microservice metrics we can both detect anomalies and provide root cause candidates without incurring the significant overheads of causal and multivariate approaches. We demonstrate that our approach can achieve top detection and localisation performance across a wide degree of different fault scenarios when compared to state-of-the-art approaches. Furthermore, we illustrate the scalability of our approach by applying it to Alibaba's extremely large real-world microservice deployment.

cs.SE↗

arXiv2025

Synergistic Motifs in Gaussian Systems

High-order interdependencies are central features of complex systems, yet a mechanistic explanation for their emergence remains elusive. Currently, it is unknown under what conditions high-order interdependencies, quantified by the information-theoretic construct of synergy, arise in systems governed by pairwise interactions. We solve this problem by providing precise sufficient and necessary conditions for when synergy prevails over low-order interdependencies in the weak interaction regime, namely, we prove that antibalanced (highly frustrated) correlational structures in Gaussian systems are sufficient for synergy-dominance and that antibalanced interaction motifs in Ornstein-Uhlenbeck processes are necessary for synergy-dominance. We validate the applicability of these analytical insights beyond the weak interaction regime, as well as in Ising, oscillatory, and empirical networks from multiple domains. Our results demonstrate that pairwise interactions can give rise to synergistic information in the absence of explicit high-order mechanisms, and highlight structural balance theory as an instrumental conceptual framework to study high-order interdependencies.

physics.soc-ph↗

arXiv2025

Complexity at Scale: A Quantitative Analysis of an Alibaba Microservice Deployment

Microservice management and testbed research often rests on assumptions about deployments that have rarely been validated at production scale. While recent studies have begun to characterise production microservice deployments, they are often limited in breadth, do not compare findings across deployments, and lack consideration of the implications of findings for commonly held assumptions. We analyse a distributed tracing dataset from Alibaba's production microservice deployment to examine its scale, heterogeneity, and dynamicity. By comparing our findings to prior measurements of Meta's MSA we illustrate both convergent and divergent properties, clarifying which patterns may generalise. Our study reveals extreme architectural scale, long-tail distributions of workloads and dependencies, highly diverse functionality, substantial call graph variability, and pronounced time-varying behaviour which diverge from assumptions underlying research models and testbeds. We summarise how these observations challenge common assumptions in research on fault management, scaling, and testbed design, and outline recommendations for more realistic future approaches and evaluations.

cs.DC↗

arXiv2024

Emergence of metastability in frustrated oscillatory networks: the key role of hierarchical modularity

Oscillatory complex networks in the metastable regime have been used to study the emergence of integrated and segregated activity in the brain, which are hypothesised to be fundamental for cognition. Yet, the parameters and the underlying mechanisms necessary to achieve the metastable regime are hard to identify, often relying on maximising the correlation with empirical functional connectivity dynamics. Here, we propose and show that the brain's hierarchically modular mesoscale structure alone can give rise to robust metastable dynamics and (metastable) chimera states in the presence of phase frustration. We construct unweighted $3$-layer hierarchical networks of identical Kuramoto-Sakaguchi oscillators, parameterized by the average degree of the network and a structural parameter determining the ratio of connections between and within blocks in the upper two layers. Together, these parameters affect the characteristic timescales of the system. Away from the critical synchronization point, we detect the emergence of metastable states in the lowest hierarchical layer coexisting with chimera and metastable states in the upper layers. Using the Laplacian renormalization group flow approach, we uncover two distinct pathways towards achieving the metastable regimes detected in these distinct layers. In the upper layers, we show how the symmetry-breaking states depend on the slow eigenmodes of the system. In the lowest layer instead, metastable dynamics can be achieved as the separation of timescales between layers reaches a critical threshold. Our results show an explicit relationship between metastability, chimera states, and the eigenmodes of the system, bridging the gap between harmonic based studies of empirical data and oscillatory models.

physics.bio-ph↗

arXiv2023

Towards inferring network properties from epidemic data

Epidemic propagation on networks represents an important departure from traditional massaction models. However, the high-dimensionality of the exact models poses a challenge to both mathematical analysis and parameter inference. By using mean-field models, such as the pairwise model (PWM), the complexity becomes tractable. While such models have been used extensively for model analysis, there is limited work in the context of statistical inference. In this paper, we explore the extent to which the PWM with the susceptible-infected-recovered (SIR) epidemic can be used to infer disease- and network-related parameters. The widely-used MLE approach exhibits several issues pertaining to parameter unidentifiability and a lack of robustness to exact knowledge about key quantities such as population size and/or proportion of under reporting. As an alternative, we considered the recently developed dynamical survival analysis (DSA). For scenarios in which there is no model mismatch, such as when data are generated via simulations, both methods perform well despite strong dependence between parameters. However, for real-world data, such as foot-and-mouth, H1N1 and COVID19, the DSA method appears more robust to potential model mismatch and the parameter estimates appear more epidemiologically plausible. Taken together, however, our findings suggest that network-based mean-field models can be used to formulate approximate likelihoods which, coupled with an efficient inference scheme, make it possible to not only learn about the parameters of the disease dynamics but also that of the underlying network.

q-bio.QM↗

arXiv2021

Probabilistic predictions of SIS epidemics on networks based on population-level observations

We predict the future course of ongoing susceptible-infected-susceptible (SIS) epidemics on regular, Erdős-Rényi and Barabási-Albert networks. It is known that the contact network influences the spread of an epidemic within a population. Therefore, observations of an epidemic, in this case at the population-level, contain information about the underlying network. This information, in turn, is useful for predicting the future course of an ongoing epidemic. To exploit this in a prediction framework, the exact high-dimensional stochastic model of an SIS epidemic on a network is approximated by a lower-dimensional surrogate model. The surrogate model is based on a birth-and-death process; the effect of the underlying network is described by a parametric model for the birth rates. We demonstrate empirically that the surrogate model captures the intrinsic stochasticity of the epidemic once it reaches a point from which it will not die out. Bayesian parameter inference allows for uncertainty about the model parameters and the class of the underlying network to be incorporated directly into probabilistic predictions. An evaluation of a number of scenarios shows that in most cases the resulting prediction intervals adequately quantify the prediction uncertainty. As long as the population-level data is available over a long-enough period, even if not sampled frequently, the model leads to excellent predictions where the underlying network is correctly identified and prediction uncertainty mainly reflects the intrinsic stochasticity of the spreading epidemic. For predictions inferred from shorter observational periods, uncertainty about parameters and network class dominate prediction uncertainty. The proposed method relies on minimal data and is numerically efficient, which makes it attractive either as a standalone inference and prediction scheme or in conjunction with other methods.

q-bio.PE↗

arXiv2020

The impact of network properties and mixing on control measures and disease-induced herd immunity in epidemic models: a mean-field model perspective

The contact structure of a population plays an important role in transmission of infection. Many ``structured models'' capture aspects of the contact structure through an underlying network or a mixing matrix. An important observation in such models, is that once a fraction $1-1/\mathcal{R}_0$ has been infected, the residual susceptible population can no longer sustain an epidemic. A recent observation of some structured models is that this threshold can be crossed with a smaller fraction of infected individuals, because the disease acts like a targeted vaccine, preferentially immunizing higher-risk individuals who play a greater role in transmission. Therefore, a limited ``first wave'' may leave behind a residual population that cannot support a second wave once interventions are lifted. In this paper, we systematically analyse a number of mean-field models for networks and other structured populations to address issues relevant to the Covid-19 pandemic. In particular, we consider herd-immunity under several scenarios. We confirm that, in networks with high degree heterogeneity, the first wave confers herd-immunity with significantly fewer infections than equivalent models with lower degree heterogeneity. However, if modelling the intervention as a change in the contact network, then this effect might become more subtle. Indeed, modifying the structure can shield highly connected nodes from becoming infected during the first wave and make the second wave more substantial. We confirm this finding by using an age-structured compartmental model parameterised with real data and comparing lockdown periods implemented either as a global scaling of the mixing matrix or age-specific structural changes. We find that results regarding herd immunity levels are strongly dependent on the model, the duration of lockdown and how lockdown is implemented.

q-bio.PE↗

arXiv2020

PDE-limits of stochastic SIS epidemics on networks

Stochastic epidemic models on networks are inherently high-dimensional and the resulting exact models are intractable numerically even for modest network sizes. Mean-field models provide an alternative but can only capture average quantities, thus offering little or no information about variability in the outcome of the exact process. In this paper we conjecture and numerically prove that it is possible to construct PDE-limits of the exact stochastic SIS epidemics on regular and Erdős-Rényi networks. To do this we first approximate the exact stochastic process at population level by a Birth-and-Death process (BD) (with a state space of $O(N)$ rather than $O(2^N)$) whose coefficients are determined numerically from Gillespie simulations of the exact epidemic on explicit networks. We numerically demonstrate that the coefficients of the resulting BD process are density-dependent, a crucial condition for the existence of a PDE limit. Extensive numerical tests for Regular and Erdős-Rényi networks show excellent agreement between the outcome of simulations and the numerical solution of the Fokker-Planck equations. Apart from a significant reduction in dimensionality, the PDE also provides the means to derive the epidemic outbreak threshold linking network and disease dynamics parameters, albeit in an implicit way. Perhaps more importantly, it enables the formulation and numerical evaluation of likelihoods for epidemic and network inference as illustrated in a worked out example.

q-bio.PE↗

arXiv2019

A new method for the robust characterisation of pairwise statistical dependency between point processes

The robust detection of statistical dependencies between the components of a complex system is a key step in gaining a network-based understanding of the system. Because of their simplicity and low computation cost, pairwise statistics are commonly used in a variety of fields. Those approaches, however, typically suffer from one or more limitations such as lack of confidence intervals requiring reliance on surrogate data, sensitivity to binning, sparsity of the signals, or short duration of the records. In this paper we develop a method for assessing pairwise dependencies in point processes that overcomes these challenges. Given two point processes $X$ and $Y$ each emitting a given number of events $m$ and $n$ in a fixed period of time $T$, we derive exact analytical expressions for the expected value and standard deviation of the number of pairs events $X_i,Y_j$ separated by a delay of less than $τ$ one should expect to observe if $X$ and $Y$ were i.i.d. uniform random variables. We prove that this statistic is normally distributed in the limit of large $T$, which enables the definition of a Z-score characterising the likelihood of the observed number of coincident events happening by chance. We numerically confirm the analytical results and show that the property of normality is robust in a wide range of experimental conditions. We then experimentally demonstrate the predictive power of the method using a noisy version of the common shock model. Our results show that our approach has excellent behaviour even in scenarios with low event density and/or when the recordings are short.

math.ST↗

arXiv2018

Mapping structural diversity in networks sharing a given degree distribution and global clustering: Adaptive resolution grid search evolution with Diophantine equation-based mutations

Methods that generate networks sharing a given degree distribution and global clustering can induce changes in structural properties other than that controlled for. Diversity in structural properties, in turn, can affect the outcomes of dynamical processes operating on those networks. Since exhaustive sampling is not possible, we propose a novel evolutionary framework for mapping this structural diversity. The three main features of this framework are: (a) subgraph-based encoding of networks, (b) exact mutations based on solving systems of Diophantine equations, and (c) heuristic diversity-driven mechanism to drive resolution changes in the MapElite algorithm. We show that our framework can elicit networks with diversity in their higher-order structure and that this diversity affects the behaviour of the complex contagion model. Through a comparison with state of the art clustered network generation methods, we demonstrate that our approach can uncover a comparably diverse range of networks without needing computationally unfeasible mixing times. Further, we suggest that the subgraph-based encoding provides greater confidence in the diversity of higher-order network structure for low numbers of samples and is the basis for explaining our results with complex contagion model. We believe that this framework could be applied to other complex landscapes that cannot be practically mapped via exhaustive sampling.

cs.SI↗

arXiv2018

Epidemic threshold in pairwise models for clustered networks: closures and fast correlations

The epidemic threshold is probably the most studied quantity in the modelling of epidemics on networks. For a large class of networks and dynamics the epidemic threshold is well studied and understood. However, it is less so for clustered networks where theoretical results are mostly limited to idealised networks. In this paper we focus on a class of models known as pairwise models where, to our knowledge, no analytical result for the epidemic threshold exists. We show that by exploiting the presence of fast variables and using some standard techniques from perturbation theory we are able to obtain the epidemic threshold analytically. We validate this new threshold by comparing it to the numerical solution of the full system. The agreement is found to be excellent over a wide range of values of the clustering coefficient, transmission rate and average degree of the network. Interestingly, we find that the analytical form of $R_0$ depends on the choice of closure, highlighting the importance of model choice when dealing with real-world epidemics. Nevertheless, we ex- pect that our method will extend to other systems in which fast variables are present.

physics.soc-ph↗

arXiv2018

Edge-based compartmental modelling of an SIR epidemic on a dual-layer static-dynamic multiplex network with tunable clustering

The duration, type and structure of connections between individuals in real-world populations play a crucial role in how diseases invade and spread. Here, we incorporate the aforementioned heterogeneities into a model by considering a dual-layer static-dynamic multiplex network. The static network layer affords tunable clustering and describes an individual's permanent community structure. The dynamic network layer describes the transient connections an individual makes with members of the wider population by imposing constant edge rewiring. We follow the edge-based compartmental modelling approach to derive equations describing the evolution of a susceptible - infected - recovered (SIR) epidemic spreading through this multiplex network of individuals. We derive the basic reproduction number, measuring the expected number of new infectious cases caused by a single infectious individual in an otherwise susceptible population. We validate model equations by showing convergence to pre-existing edge-based compartmental model equations in limiting cases and by comparison with stochastically simulated epidemics. We explore the effects of altering model parameters and multiplex network attributes on resultant epidemic dynamics. We validate the basic reproduction number by plotting its value against associated final epidemic sizes measured from simulation and predicted by model equations for a number of setups. Further, we explore the effect of varying individual model parameters on the basic reproduction number. We conclude with a discussion of the significance and interpretation of the model and its relation to existing research literature. We highlight intrinsic limitations and potential extensions of the present model and outline future research considerations, both experimental and theoretical.

physics.soc-ph↗

arXiv2018

Inferring the time-varying functional connectivity of large-scale computer networks from emitted events

We consider the problem of inferring the functional connectivity of a large-scale computer network from sparse time series of events emitted by its nodes. We do so under the following three domain-specific constraints: (a) non-stationarity of the functional connectivity due to unknown temporal changes in the network, (b) sparsity of the time-series of events that limits the effectiveness of classical correlation-based analysis, and (c) lack of an explicit model describing how events propagate through the network. Under the assumption that the probability of two nodes being functionally connected correlates with the mean delay between their respective events, we develop an inference method whose output is an undirected weighted network where the weight of an edge between two nodes denotes the probability of these nodes being functionally connected. Using a combination of windowing and convolution to calculate at each time window a score quantifying the likelihood of a pair of nodes emitting events in quick succession, we develop a model of time-varying connectivity whose parameters are determined by maximising the model's predictive power from one time window to the next. To assess the effectiveness of our inference method, we construct synthetic data for which ground truth is available and use these data to benchmark our approach against three state-of-the-art inference methods. We conclude by discussing its application to data from a real-world large-scale computer network.

cs.LG↗

arXiv2015

Generation and analysis of networks with a prescribed degree sequence and subgraph family: Higher-order structure matters

Designing algorithms that generate networks with a given degree sequence while varying both subgraph composition and distribution of subgraphs around nodes is an important but challenging research problem. Current algorithms lack control of key network parameters, the ability to specify to what subgraphs a node belongs to, come at a considerable complexity cost or, critically, sample from a limited ensemble of networks. To enable controlled investigations of the impact and role of subgraphs, especially for epidemics, neuronal activity or complex contagion, it is essential that the generation process be versatile and the generated networks as diverse as possible. In this paper, we present two new network generation algorithms that use subgraphs as building blocks to construct networks preserving a given degree sequence. Additionally, these algorithms provide control over clustering both at node and global level. In both cases, we show that, despite being constrained by a degree sequence and global clustering, generated networks have markedly different topologies as evidenced by both subgraph prevalence and distribution around nodes, and large-scale network structure metrics such as path length and betweenness measures. Simulations of standard epidemic and complex contagion models on those networks reveal that degree distribution and global clustering do not always accurately predict the outcome of dynamical processes taking place on them. We conclude by discussing the benefits and limitations of both methods.

physics.soc-ph↗

arXiv2015

Using novelty-biased GA to sample diversity in graphs satisfying constraints

The structure of the network underlying many complex systems, whether artificial or natural, plays a significant role in how these systems operate. As a result, much emphasis has been placed on accurately describing networks using network theoretic metrics. When it comes to generating networks with similar properties, however, the set of available techniques and properties that can be controlled for remains limited. Further, whilst it is becoming clear that some of the metrics currently used to control the generation of such networks are not very prescriptive so that networks could potentially exhibit very different higher-order structure within those constraints, network generating algorithms typically produce fairly contrived networks and lack mechanisms by which to systematically explore the space of network solutions. In this paper, we explore the potential of a multi-objective novelty-biased GA to provide a viable alternative to these algorithms. We believe our results provide the first proof of principle that (i) it is possible to use GAs to generate graphs satisfying set levels of key classical graph theoretic properties and (ii) it is possible to generate diverse solutions within these constraints. The paper is only a preliminary step, however, and we identify key avenues for further development.

physics.soc-ph↗

arXiv2014

Beyond clustering: Mean-field dynamics on networks with arbitrary subgraph composition

Clustering is the propensity of nodes that share a common neighbour to be connected. It is ubiquitous in many networks but poses many modelling challenges. Clustering typically manifests itself by a higher than expected frequency of triangles, and this has led to the principle of constructing networks from such building blocks. This approach has been generalised to networks being constructed from a set of more exotic subgraphs. As long as these are fully connected, it is then possible to derive mean-field models that approximate epidemic dynamics well. However, there are virtually no results for non-fully connected subgraphs. In this paper, we provide a general and automated approach to deriving a set of ordinary differential equations, or mean-field model, that describes, to a high degree of accuracy, the expected values of system-level quantities, such as the prevalence of infection. Our approach offers a previously unattainable degree of control over the arrangement of subgraphs and network characteristics such as classical node degree, variance and clustering. The combination of these features makes it possible to generate families of networks with different subgraph compositions while keeping classical network metrics constant. Using our approach, we show that higher-order structure realised either through the introduction of loops of different sizes or by generating clustered networks based on different subgraphs, leads to significant differences in epidemic dynamics despite controlling for basic network metrics.

math.DS↗