arXiv subjects

Weisi Guo

Publications and source records attributed to Weisi Guo.

At least 73 records · Page 4Linked to original sources

arXiv2020

Graph Hierarchy: A novel approach to understanding hierarchical structures in complex networks

Trophic coherence, a measure of a graph's hierarchical organisation, has been shown to be linked to a graph's structural and dynamical aspects such as cyclicity, stability and normality. Trophic levels of vertices can reveal their functional properties and partition and rank the vertices accordingly. Yet trophic levels and hence trophic coherence can only be defined on graphs with basal vertices, vertices with zero in-degree. Consequently, trophic analysis of graphs had been restricted until now. In this paper we introduce a novel framework, a generalisation of trophic levels, which we call hierarchical levels, that can be defined on any simple graph. Within this general framework, we develop additional metrics named influence centrality, a measure of a vertices ability to influence dynamics, and democracy coefficient, a measure of overall feedback in the system, both of which have implications for the controllability of complex systems. We discuss how our generalisation relates to previous attempts and what new insights are illuminated on the topological and dynamical aspects of graphs. Finally, we show how the hierarchical structure of a network relates to the incidence rate in a SIS epidemic model.

physics.soc-ph↗

arXiv2020

Deep Learning Methods for Solving Linear Inverse Problems: Research Directions and Paradigms

The linear inverse problem is fundamental to the development of various scientific areas. Innumerable attempts have been carried out to solve different variants of the linear inverse problem in different applications. Nowadays, the rapid development of deep learning provides a fresh perspective for solving the linear inverse problem, which has various well-designed network architectures results in state-of-the-art performance in many applications. In this paper, we present a comprehensive survey of the recent progress in the development of deep learning for solving various linear inverse problems. We review how deep learning methods are used in solving different linear inverse problems, and explore the structured neural network architectures that incorporate knowledge used in traditional methods. Furthermore, we identify open challenges and potential future directions along this research line.

eess.SP↗

arXiv2020

Scalable Partial Explainability in Neural Networks via Flexible Activation Functions

Achieving transparency in black-box deep learning algorithms is still an open challenge. High dimensional features and decisions given by deep neural networks (NN) require new algorithms and methods to expose its mechanisms. Current state-of-the-art NN interpretation methods (e.g. Saliency maps, DeepLIFT, LIME, etc.) focus more on the direct relationship between NN outputs and inputs rather than the NN structure and operations itself. In current deep NN operations, there is uncertainty over the exact role played by neurons with fixed activation functions. In this paper, we achieve partially explainable learning model by symbolically explaining the role of activation functions (AF) under a scalable topology. This is carried out by modeling the AFs as adaptive Gaussian Processes (GP), which sit within a novel scalable NN topology, based on the Kolmogorov-Arnold Superposition Theorem (KST). In this scalable NN architecture, the AFs are generated by GP interpolation between control points and can thus be tuned during the back-propagation procedure via gradient descent. The control points act as the core enabler to both local and global adjustability of AF, where the GP interpolation constrains the intrinsic autocorrelation to avoid over-fitting. We show that there exists a trade-off between the NN's expressive power and interpretation complexity, under linear KST topology scaling. To demonstrate this, we perform a case study on a binary classification dataset of banknote authentication. By quantitatively and qualitatively investigating the mapping relationship between inputs and output, our explainable model can provide interpretation over each of the one-dimensional attributes. These early results suggest that our model has the potential to act as the final interpretation layer for deep neural networks.

cs.LG↗

arXiv2020

Uncertainty of Resilience in Complex Networks with Nonlinear Dynamics

Resilience is a system's ability to maintain its function when perturbations and errors occur. Whilst we understand low-dimensional networked systems' behavior well, our understanding of systems consisting of a large number of components is limited. Recent research in predicting the network level resilience pattern has advanced our understanding of the coupling relationship between global network topology and local nonlinear component dynamics. However, when there is uncertainty in the model parameters, our understanding of how this translates to uncertainty in resilience is unclear for a large-scale networked system. Here we develop a polynomial chaos expansion method to estimate the resilience for a wide range of uncertainty distributions. By applying this method to case studies, we not only reveal the general resilience distribution with respect to the topology and dynamics sub-models, but also identify critical aspects to inform better monitoring to reduce uncertainty.

eess.SY↗

arXiv2020

Neural Network Approximation of Graph Fourier Transforms for Sparse Sampling of Networked Flow Dynamics

Infrastructure monitoring is critical for safe operations and sustainability. Water distribution networks (WDNs) are large-scale networked critical systems with complex cascade dynamics which are difficult to predict. Ubiquitous monitoring is expensive and a key challenge is to infer the contaminant dynamics from partial sparse monitoring data. Existing approaches use multi-objective optimisation to find the minimum set of essential monitoring points, but lack performance guarantees and a theoretical framework. Here, we first develop Graph Fourier Transform (GFT) operators to compress networked contamination spreading dynamics to identify the essential principle data collection points with inference performance guarantees. We then build autoencoder (AE) inspired neural networks (NN) to generalize the GFT sampling process and under-sample further from the initial sampling set, allowing a very small set of data points to largely reconstruct the contamination dynamics over real and artificial WDNs. Various sources of the contamination are tested and we obtain high accuracy reconstruction using around 5-10% of the sample set. This general approach of compression and under-sampled recovery via neural networks can be applied to a wide range of networked infrastructures to enable digital twins.

cs.LG↗

arXiv2019

Gang Confrontation: The case of Medellin (Colombia)

Protracted conflict is one of the largest human challenges that have persistently undermined economic and social progress. In recent years, there has been increased emphasis on using statistical and physical science models to better understand both the universal patterns and the underlying mechanics of conflict. Whilst macroscopic power-law fractal patterns have been shown for death-toll in wars and self-excitation models have been shown for roadside ambush attacks, very few works deal with the challenge of complex dynamics between gangs at the intra-city scale. Here, based on contributions to the historical memory of the conflict in Colombia, Medellin's gang-confrontation-network is presented. It is shown that socio-economic and violence indexes are moderate to highly correlated to the structure of the network. Specifically, the death-toll of conflict is strongly influenced by the leading eigenvalues of the gangs' conflict adjacency matrix, which serves a proxy for unstable self-excitation from revenge attacks. The distribution of links based on the geographic distance between gangs in confrontation leads to the confirmation that territorial control is a main catalyst of violence and retaliation among gangs. Additionally, the Boltzmann-Lotka-Volterra (BLV) dynamic interaction network analysis is applied to quantify the spatial embeddedness of the dynamic relationship between conflicting gangs in Medellin, results suggest that more involved and comprehensive models are needed to described the dynamics of Medellin's armed conflict.

physics.soc-ph↗

arXiv2019

Dynamic Complex Network Analysis of PM2.5 Concentrations in the UK using Hierarchical Directed Graphs

Worldwide exposure to fine atmospheric particles can exasperate the risk of a wide range of heart and respiratory diseases, due to their ability to penetrate deep into the lungs and blood streams. Epidemiological studies in Europe and elsewhere have established the evidence base pointing to the important role of PM2.5 in causing over 4 million deaths per year. Traditional approaches to model atmospheric transportation of particles suffer from high dimensionality from both transport and chemical reaction processes, making multi-sale causal inference challenging. We apply alternative model reduction methods: a data-driven directed graph representation to infer spatial embeddedness and causal directionality. Using PM2.5 concentrations in 14 UK cities over a 12 month period, we construct an undirected correlation and a directed Granger causality network. We show for both reduced-order cases, the UK is divided into two a northern and southern connected city communities, with greater spatial embedding in spring and summer. We go on to infer stability to disturbances via the network trophic coherence parameter, whereby we found that winter had the greatest vulnerability. As a result of our novel graph-based reduced modeling, we are able to represent high-dimensional knowledge into a causal inference and stability framework.

physics.ao-ph↗

arXiv2019

Explainable Artificial Intelligence (XAI) for 6G: Improving Trust between Human and Machine

As the 5th Generation (5G) mobile networks are bringing about global societal benefits, the design phase for the 6th Generation (6G) has started. 6G will need to enable greater levels of autonomy, improve human machine interfacing, and achieve deep connectivity in more diverse environments. The need for increased explainability to enable trust is critical for 6G as it manages a wide range of mission critical services (e.g. autonomous driving) to safety critical tasks (e.g. remote surgery). As we migrate from traditional model-based optimisation to deep learning, the trust we have in our optimisation modules decrease. This loss of trust means we cannot understand the impact of: 1) poor/bias/malicious data, and 2) neural network design on decisions; nor can we explain to the engineer or the public the network's actions. In this review, we outline the core concepts of Explainable Artificial Intelligence (XAI) for 6G, including: public and legal motivations, definitions of explainability, performance vs. explainability trade-offs, methods to improve explainability, and frameworks to incorporate XAI into future wireless systems. Our review is grounded in cases studies for both PHY and MAC layer optimisation, and provide the community with an important research area to embark upon.

eess.SP↗

arXiv2019

Forecasting Wireless Demand with Extreme Values using Feature Embedding in Gaussian Processes

Wireless traffic prediction is a fundamental enabler to proactive network optimisation in beyond 5G. Forecasting extreme demand spikes and troughs due to traffic mobility is essential to avoiding outages and improving energy efficiency. Current state-of-the-art deep learning forecasting methods predominantly focus on overall forecast performance and do not offer probabilistic uncertainty quantification (UQ). Whilst Gaussian Process (GP) models have UQ capability, it is not able to predict extreme values very well. Here, we design a feature embedding (FE) kernel for a GP model to forecast traffic demand with extreme values. Using real 4G base station data, we compare our FE-GP performance against both conventional naive GPs, ARIMA models, as well as demonstrate the UQ output. For short-term extreme value prediction, we demonstrated a 32\% reduction vs. S-ARIMA and 17\% reduction vs. Naive-GP. For long-term average value prediction, we demonstrated a 21\% reduction vs. S-ARIMA and 12\% reduction vs. Naive-GP. The FE kernel also enabled us to create a flexible trade-off between overall forecast accuracy against peak-trough accuracy. The advantage over neural network (e.g. CNN, LSTM) is that the probabilistic forecast uncertainty can inform us of the risk of predictions, as well as the full posterior distribution of the forecast.

eess.SP↗

arXiv2019

Green Deep Reinforcement Learning for Radio Resource Management: Architecture, Algorithm Compression and Challenge

AI heralds a step-change in the performance and capability of wireless networks and other critical infrastructures. However, it may also cause irreversible environmental damage due to their high energy consumption. Here, we address this challenge in the context of 5G and beyond, where there is a complexity explosion in radio resource management (RRM). On the one hand, deep reinforcement learning (DRL) provides a powerful tool for scalable optimization for high dimensional RRM problems in a dynamic environment. On the other hand, DRL algorithms consume a high amount of energy over time and risk compromising progress made in green radio research. This paper reviews and analyzes how to achieve green DRL for RRM via both architecture and algorithm innovations. Architecturally, a cloud based training and distributed decision-making DRL scheme is proposed, where RRM entities can make lightweight deep local decisions whilst assisted by on-cloud training and updating. On the algorithm level, compression approaches are introduced for both deep neural networks and the underlying Markov Decision Processes, enabling accurate low-dimensional representations of challenges. To scale learning across geographic areas, a spatial transfer learning scheme is proposed to further promote the learning efficiency of distributed DRL entities by exploiting the traffic demand correlations. Together, our proposed architecture and algorithms provide a vision for green and on-demand DRL capability.

cs.LG↗

arXiv2019

Nonlinear Dynamic Models of Conflict via Multiplexed Interaction Networks

The risk of conflict is exasperated by a multitude of internal and external factors. Current multivariate analysis paints diverse causal risk profiles that vary with time. However, these profiles evolve and a universal model to understand that evolution remains absent. Most of the current conflict analysis is data-driven and conducted at the individual country or region level, often in isolation. Consistent consideration of multi-scale interactions and their non-linear dynamics is missing. Here, we develop a multiplexed network model, where each city is modelled as a non-linear bi-stable system with stable states in either war or peace. The causal factor categories which exasperate the risk of conflict are each modelled as a network layer. We consider 3 layers: (1) core geospatial network of interacting cities reflecting ground level interactions, (2) cultural network of interacting countries reflecting cultural groupings, and (3) political network of interacting countries reflecting alliances. Together, they act as drivers to push cities towards or pull cities away from war. Using a variety of data sources relative to 2002-2016, we show, that our model correctly predicts the transitions from war to peace and peace to war with F1 score of 0.78 to 0.92 worldwide at the city scale resolution. As many conflicts during this period are auto-regressive (e.g. the War on Terror in Afghanistan and Iraq, the Narco War across the Americas), we can predict the emergence of new war or new peace. We demonstrate successful predictions across a wide range of conflict genres and we perform causal discovery by identifying which model component led to the correct prediction. In the cases of Somalia (2008-13), Myanmar (2013-15), Colombia (2011-14), Libya (2014-16), and Yemen (2011-13) we identify the set of most likely causal factors and how it may differ across a country and change over time.

physics.soc-ph↗

arXiv2019

Assessing Simulations of Imperial Dynamics and Conflict in the Ancient World

The development of models to capture large-scale dynamics in human history is one of the core contributions of cliodynamics. Most often, these models are assessed by their predictive capability on some macro-scale and aggregated measure and compared to manually curated historical data. In this report, we consider the model from Turchin et al. (2013), where the evaluation is done on the prediction of "imperial density": the relative frequency with which a geographical area belonged to large-scale polities over a certain time window. We implement the model and release both code and data for reproducibility. We then assess its behaviour against three historical data sets: the relative size of simulated polities vs historical ones; the spatial correlation of simulated imperial density with historical population density; the spatial correlation of simulated conflict vs historical conflict. At the global level, we show good agreement with population density ($R^2 < 0.75$), and some agreement with historical conflict in Europe ($R^2 < 0.42$). The model instead fails to reproduce the historical shape of individual polities. Finally, we tweak the model to behave greedily by having polities preferentially attacking weaker neighbours. Results significantly degrade, suggesting that random attacks are a key trait of the original model. We conclude by proposing a way forward by matching the probabilistic imperial strength from simulations to inferred networked communities from real settlement data.

physics.soc-ph↗

arXiv2019

Molecular Information Delivery in Porous Media

Information delivery via molecular signals is abundant in nature and potentially useful for industry sensing. Many propagation channels (e.g., tissue membranes and catalyst beds) contain porous medium materials and the impact this has on communication performance is not well understood. Here, communication through realistic porous channels is investigated for the first time via statistical breakthrough curves. Assuming that the number of arrived molecules can be approximated as a Gaussian random variable and using fully resolved computational fluid dynamics results for the breakthrough curves, the numerical results for the throughput, mutual information, error probability, and information diversity gain are presented. Using these numerical results, the unique characteristics of the porous medium channel are revealed.

cs.IT↗

arXiv2019

Quantifying Networked Resilience via Multi-Scale Feedback Loops in Water Distribution Networks

Water distribution networks (WDNs) are one of the most important man-made infrastructures. Resilience, the ability to respond to disturbances and recover to a desirable state, is of vital importance to our society. There is increasing evidence that the resilience of networked infrastructures with dynamic signals depends on their network topological properties. Whilst existing literature has examined a variety of complex network centrality and spectral properties, very little attention has been given to understand multi-scale flow-based feedback loops and its impact on overall stability. Here, resilience arising from multi-scale feedback loops is inferred using a proxy measure called trophic coherence. This is hypothesised to have a strong impact on both the pressure deficit and the water age. Our results show that trophic coherence is positively correlated with the time to recovery (0.62 to 0.52), while it is negatively correlated with the diffusion of a disruption (-0.66 to -0.52). Finally, we apply random forest analysis to combine resilience measures, showing that the new resilience ensemble provides a more accurate measure for networked resilience.

physics.soc-ph↗

arXiv2019

Optimal Sampling for Dynamic Complex Networks with Graph-Bandlimited Initialization

Many engineering, social, and biological complex systems consist of dynamical elements connected via a large-scale network. Monitoring the network's dynamics is essential for a variety of maintenance and scientific purposes. Whilst we understand how to optimally sample a single dynamic element or a non-dynamic graph, we do not possess a theory on how to optimally sample networked dynamical elements. Here, we study nonlinear dynamic graph signals on a fixed complex network. We define the necessary conditions for optimal sampling in the combining time- and graph-domain to fully recover the networked dynamics. We firstly interpret the networked dynamics into a linearized matrix. Then, we prove that the dynamic signals can be sampled and fully recovered if the networked dynamics is stable and their initialization is bandlimited in the graph spectral domain. This new theory directly maps optimal sampling locations and rates to the graph properties and governing nonlinear dynamics. This can inform the optimal placement of experimental probes and sensors on dynamical networks as well as inform the design of each sensor's optimal sampling rate. We motivate the reader with two examples of recovering the networked dynamics for: social population growth and networked protein biochemical interactions with both bandlimited and arbitrary initialization.

eess.SP↗

arXiv2019

Common Statistical Patterns in Urban Terrorism

The underlying reasons behind modern terrorism are seemingly complex and intangible. Despite diverse causal mechanisms, research has shown that there exists general statistical patterns at the global scale that can shed light on human confrontation behaviour. Whilst many policing and counter-terrorism operations are conducted at a city level, there has been a lack of research in building city-level resolution prediction engines based on statistical patterns. For the first time, the paper shows that there exists general commonalities between global cities under terrorist attacks. By examining over 30,000 geo-tagged terrorism acts over 7000 cities worldwide from 2002 to today, the results shows the following. All cities experience attacks $A$ that are uncorrelated to the population and separated by a time interval $t$ that is negative exponentially distributed $\sim \exp(-A^{-1})$, with a death-toll per attack that follows a power law distribution. The prediction parameters yield a high confidence of explaining up to 87\% of the variations in frequency and 89\% in the death-toll data. These findings show that the aggregate statistical behaviour of terror attacks are seemingly random and memoryless for all global cities. The enabled the author to develop a data-driven city-specific prediction system and we quantify its information theoretic uncertainty and information loss. Further analysis show that there appears to be an increase in the uncertainty over the predictability of attacks, challenging our ability to develop effective counter-terrorism capabilities.

stat.AP↗

arXiv2019

Optimal Sampling of Water Distribution Network Dynamics using Graph Fourier Transform

Water Distribution Networks (WDNs) are critical infrastructures that ensure safe drinking water. One of the major threats is the accidental or intentional injection of pollutants. Data collection remains challenging in underground WDNs and in order to quantify its threat to end users, modeling pollutant spread with minimal sensor data is can important open challenge. Existing approaches using numerical optimisation suffer from scalability issues and lack detailed insight and performance guarantees. Applying general data-driven approaches such as compressed sensing (CS) offer limited improvements in sample node reduction. Graph theoretic approaches link topology (e.g. Laplacian spectra) to optimal sensing locations, it neglects the complex dynamics. In this work, we introduce a novel Graph Fourier Transform (GFT) that exploits the low-rank property to optimally sample junction nodes in WDNs. The proposed GFT allows us to fully recover the full network dynamics using a subset of data sampled at the identified nodes. The proposed GFT technique offers attractive improvements over existing numerical optimisation, compressed sensing, and graph theoretic approaches. Our results show that, on average, with nearly 30-40\% of the junctions monitored, we are able to fully recover the dynamics of the whole network. The framework is useful beyond the application of WDNs and can be applied to a variety of infrastructure sensing for digital twin modeling.

cs.IT↗

arXiv2019

Probabilistic Stability of Traffic Load Balancing on Wireless Complex Networks

Load balancing between adjacent base stations (BSs) is important for balancing load distributions and improving service provisioning. Whilst load balancing between any given pair of BSs is beneficial, cascade load sharing can cause network level instability that is hard to predict. The relationship between each BS's load balancing dynamics and the network topology is not understood. In this seminal work on stability analysis, we consider a frequency re-use network with no interference, whereby load balancing dynamics doesn't perturb the individual cells' capacity. Our novelty is to show an exact analytical and also a probabilistic relationship for stability, relating generalized local load balancing dynamics with generalized network topology, as well as the uncertainty we have in load balancing parameters. We prove that the stability analysis given is valid for any generalized load balancing dynamics and topological cell deployment and we believe this general relationship can inform the joint design of both the load balancing dynamics and the neighbour list of the network.

cs.NI↗