arXiv subjects

Nima Dehmamy

Publications and source records attributed to Nima Dehmamy.

24 records · Page 2Linked to original sources

arXiv2020

Finding Patient Zero: Learning Contagion Source with Graph Neural Networks

Locating the source of an epidemic, or patient zero (P0), can provide critical insights into the infection's transmission course and allow efficient resource allocation. Existing methods use graph-theoretic centrality measures and expensive message-passing algorithms, requiring knowledge of the underlying dynamics and its parameters. In this paper, we revisit this problem using graph neural networks (GNNs) to learn P0. We establish a theoretical limit for the identification of P0 in a class of epidemic models. We evaluate our method against different epidemic models on both synthetic and a real-world contact network considering a disease with history and characteristics of COVID-19. % We observe that GNNs can identify P0 close to the theoretical bound on accuracy, without explicit input of dynamics or its parameters. In addition, GNN is over 100 times faster than classic methods for inference on arbitrary graph topologies. Our theoretical bound also shows that the epidemic is like a ticking clock, emphasizing the importance of early contact-tracing. We find a maximum time after which accurate recovery of the source becomes impossible, regardless of the algorithm used.

cs.SI↗

arXiv2019

Understanding the Representation Power of Graph Neural Networks in Learning Graph Topology

To deepen our understanding of graph neural networks, we investigate the representation power of Graph Convolutional Networks (GCN) through the looking glass of graph moments, a key property of graph topology encoding path of various lengths. We find that GCNs are rather restrictive in learning graph moments. Without careful design, GCNs can fail miserably even with multiple layers and nonlinear activation functions. We analyze theoretically the expressiveness of GCNs, concluding a modular GCN design, using different propagation rules with residual connections could significantly improve the performance of GCN. We demonstrate that such modular designs are capable of distinguishing graphs from different graph generation models for surprisingly small graphs, a notoriously difficult problem in network science. Our investigation suggests that, depth is much more influential than width, with deeper GCNs being more capable of learning higher order graph moments. Additionally, combining GCN modules with different propagation rules is critical to the representation power of GCNs.

cs.LG↗

arXiv2018

Separation of time scales and direct computation of weights in deep neural networks

Artificial intelligence is revolutionizing our lives at an ever increasing pace. At the heart of this revolution is the recent advancements in deep neural networks (DNN), learning to perform sophisticated, high-level tasks. However, training DNNs requires massive amounts of data and is very computationally intensive. Gaining analytical understanding of the solutions found by DNNs can help us devise more efficient training algorithms, replacing the commonly used mthod of stochastic gradient descent (SGD). We analyze the dynamics of SGD and show that, indeed, direct computation of the solutions is possible in many cases. We show that a high performing setup used in DNNs introduces a separation of time-scales in the training dynamics, allowing SGD to train layers from the lowest (closest to input) to the highest. We then show that for each layer, the distribution of solutions found by SGD can be estimated using a class-based principal component analysis (PCA) of the layer's input. This finding allows us to forgo SGD entirely and directly derive the DNN parameters using this class-based PCA, which can be well estimated using significantly less data than SGD. We implement these results on image datasets MNIST, CIFAR10 and CIFAR100 and find that, in fact, layers derived using our class-based PCA perform comparable or superior to neural networks of the same size and architecture trained using SGD. We also confirm that the class-based PCA often converges using a fraction of the data required for SGD. Thus, using our method training time can be reduced both by requiring less training data than SGD, and by eliminating layers in the costly backpropagation step of the training.

cs.LG↗

arXiv2016

Crises and Physical Phases of a Bipartite Market Model

We analyze the linear response of a market network to shocks based on the bipartite market model we introduced in an earlier paper, which we claimed to be able to identify the time-line of the 2009-2011 Eurozone crisis correctly. We show that this model has three distinct phases that can broadly be categorized as "stable" and "unstable". Based on the interpretation of our behavioral parameters, the stable phase describes periods where investors and traders have confidence in the market (e.g. predict that the market rebounds from a loss). We show that the unstable phase happens when there is a lack of confidence and seems to describe "boom-bust" periods in which changes in prices are exponential. We analytically derive these phases and where the phase transition happens using a mean field approximation of the model. We show that the condition for stability is $αβ<1$ with $α$ being the inverse of the "price elasticity" and $β$ the "income elasticity of demand", which measures how rash the investors make decisions. We also show that in the mean-field limit this model reduces to the Langevin model by Bouchaud et al. for price returns.

q-fin.RM↗

arXiv2015

Arbitrary degree distribution and high clustering in networks of locally interacting agents

Many real world networks, such as social networks, are primarily formed through local interactions between agents. Additionally, in contrast with common network models, social and biological networks exhibit a high degree of clustering. Here we construct a class of network growth models based on local interactions on a metric space, capable of producing arbitrary degree distributions as well as a naturally high degree of clustering akin to biological networks. As a specific example, we study the case of random- walking agents, though most results hold for any linear stochastic dynamics. Agents form bonds when they meet at designated locations we refer to as "rendezvous points." The spatial distribution of the rendezvous points determines key characteristics of the network such as the degree distribution. For any arbitrary (monotonic) degree distribution, we are able to analytically solve for the required rendezvous point distribution.

physics.soc-ph↗

arXiv2014

Classical mechanics of economic networks

Financial networks are dynamic. To assess their systemic importance to the world-wide economic network and avert losses we need models that take the time variations of the links and nodes into account. Using the methodology of classical mechanics and Laplacian determinism we develop a model that can predict the response of the financial network to a shock. We also propose a way of measuring the systemic importance of the banks, which we call BankRank. Using European Bank Authority 2011 stress test exposure data, we apply our model to the bipartite network connecting the largest institutional debt holders of the troubled European countries (Greece, Italy, Portugal, Spain, and Ireland). From simulating our model we can determine whether a network is in a "stable" state in which shocks do not cause major losses, or a "unstable" state in which devastating damages occur. Fitting the parameters of the model, which play the role of physical coupling constants, to Eurozone crisis data shows that before the Eurozone crisis the system was mostly in a "stable" regime, and that during the crisis it transitioned into an "unstable" regime. The numerical solutions produced by our model match closely the actual time-line of events of the crisis. We also find that, while the largest holders are usually more important, in the unstable regime smaller holders also exhibit systemic importance. Our model also proves useful for determining the vulnerability of banks and assets to shocks. This suggests that our model may be a useful tool for simulating the response dynamics of shared portfolio networks.

q-fin.RM↗