Preventing the spread of infectious diseases requires implementing interventions at various levels of government and evaluating the potential impact and efficacy of those preemptive measures. Agent-based modeling can be used for detailed studies of the spread of such diseases in the presence of possible interventions. The computational cost of modeling epidemic diffusion through large social contact networks necessitates the use of parallel algorithms and resources in order to achieve quick turnaround times. In this work, we present Loimos, a scalable parallel framework for simulating epidemic diffusion. Loimos uses a hybrid of time-stepping and discrete event simulation to model disease spread, and is implemented on top of Charm++, an asynchronous, many-task runtime that enables over-decomposition and adaptive overlap of computation and communication. We demonstrate that Loimos is able to achieve significant speedups while scaling to large core counts. In particular, Loimos is able to simulate 200 days of a COVID19 outbreak on a digital twin of California in about 42 seconds, for an average of 4.6 billion traversed edges per second (TEPS), using 4096 cores on Perlmutter at NERSC.

Causal effect estimation on a graph of connected units is often complicated by entangled treatments, where the treatment assignment is not independent for each individual. This presents multiple challenges: accurately modeling treatment assignment mechanisms, adjusting for both observed and unobserved confounders to mitigate confounding bias, and constructing instrumental variables to adjust unobserved confounders within a graph structure. Prior research on estimating the causal effects of entangled treatments either assumed no unobserved confounders or relied on the manual selection of IVs, leading to gaps in the methodology. To bridge these gaps and build upon previous work, we introduce the Graph-Disentanglement Instrumental Variable (GDIV) model, a novel approach employing both Graph Neural Networks (GNNs) and Adversarial Networks to assess the causal effects on nodes in a graph, considering observed/unobserved confounders and the intricacies of treatment entanglement. Our GDIV estimator is validated through extensive experiments across synthetic and semisynthetic datasets, demonstrating its better performance over state-of-the-art methods. The ablation studies and robustness experiments verify the benefits of leveraging adversarial networks to generate IVs that satisfy the required assumptions.

We consider the setting of cascades that result from contagion dynamics on large realistic contact networks. We address the question of whether the structural properties of a (partially) observed cascade can characterize the contagion scenario and identify the interventions that might be in effect. Using epidemic spread as a concrete example, we study how social interventions such as compliance in social distancing, extent (and efficacy) of vaccination, and the transmissibility of disease can be inferred. The techniques developed are more generally applicable to other contagions as well.

Infectious disease forecasting has been a key focus in the recent past owing to the COVID-19 pandemic and has proved to be an important tool in controlling the pandemic. With the advent of reliable spatiotemporal data, graph neural network models have been able to successfully model the inter-relation between the cross-region signals to produce quality forecasts, but like most deep-learning models they do not explicitly incorporate the underlying causal mechanisms. In this work, we employ a causal mechanistic model to guide the learning of the graph embeddings and propose a novel learning framework -- Causal-based Graph Neural Network (CausalGNN) that learns spatiotemporal embedding in a latent space where graph input features and epidemiological context are combined via a mutually learning mechanism using graph-based non-linear transformations. We design an attention-based dynamic GNN module to capture spatial and temporal disease dynamics. A causal module is added to the framework to provide epidemiological context for node embedding via ordinary differential equations. Extensive experiments on forecasting daily new cases of COVID-19 at global, US state, and US county levels show that the proposed method outperforms a broad range of baselines. The learned model which incorporates epidemiological context organizes the embedding in an efficient way by keeping the parameter size small leading to robust and accurate forecasting performance across various datasets.

We study allocation of COVID-19 vaccines to individuals based on the structural properties of their underlying social contact network. Using a realistic representation of a social contact network for the Commonwealth of Virginia, we study how a limited number of vaccine doses can be strategically distributed to individuals to reduce the overall burden of the pandemic. We show that allocation of vaccines based on individuals' degree (number of social contacts) and total social proximity time is significantly more effective than the usually used age-based allocation strategy in reducing the number of infections, hospitalizations and deaths. The overall strategy is robust even: (i) if the social contacts are not estimated correctly; (ii) if the vaccine efficacy is lower than expected or only a single dose is given; (iii) if there is a delay in vaccine production and deployment; and (iv) whether or not non-pharmaceutical interventions continue as vaccines are deployed. For reasons of implementability, we have used degree, which is a simple structural measure and can be easily estimated using several methods, including the digital technology available today. These results are significant, especially for resource-poor countries, where vaccines are less available, have lower efficacy, and are more slowly distributed.

We study the role of vaccine acceptance in controlling the spread of COVID-19 in the US using AI-driven agent-based models. Our study uses a 288 million node social contact network spanning all 50 US states plus Washington DC, comprised of 3300 counties, with 12.59 billion daily interactions. The highly-resolved agent-based models use realistic information about disease progression, vaccine uptake, production schedules, acceptance trends, prevalence, and social distancing guidelines. Developing a national model at this resolution that is driven by realistic data requires a complex scalable workflow, model calibration, simulation, and analytics components. Our workflow optimizes the total execution time and helps in improving overall human productivity.This work develops a pipeline that can execute US-scale models and associated workflows that typically present significant big data challenges. Our results show that, when compared to faster and accelerating vaccinations, slower vaccination rates due to vaccine hesitancy cause averted infections to drop from 6.7M to 4.5M, and averted total deaths to drop from 39.4K to 28.2K nationwide. This occurs despite the fact that the final vaccine coverage is the same in both scenarios. Improving vaccine acceptance by 10% in all states increases averted infections from 4.5M to 4.7M (a 4.4% improvement) and total deaths from 28.2K to 29.9K (a 6% increase) nationwide. The analysis also reveals interesting spatio-temporal differences in COVID-19 dynamics as a result of vaccine acceptance. To our knowledge, this is the first national-scale analysis of the effect of vaccine acceptance on the spread of COVID-19, using detailed and realistic agent-based models.

The COVID-19 global outbreak represents the most significant epidemic event since the 1918 influenza pandemic. Simulations have played a crucial role in supporting COVID-19 planning and response efforts. Developing scalable workflows to provide policymakers quick responses to important questions pertaining to logistics, resource allocation, epidemic forecasts and intervention analysis remains a challenging computational problem. In this work, we present scalable high performance computing-enabled workflows for COVID-19 pandemic planning and response. The scalability of our methodology allows us to run fine-grained simulations daily, and to generate county-level forecasts and other counterfactual analysis for each of the 50 states (and DC), 3140 counties across the USA. Our workflows use a hybrid cloud/cluster system utilizing a combination of local and remote cluster computing facilities, and using over 20,000 CPU cores running for 6-9 hours every day to meet this objective. Our state (Virginia), state hospital network, our university, the DOD and the CDC use our models to guide their COVID-19 planning and response efforts. We began executing these pipelines March 25, 2020, and have delivered and briefed weekly updates to these stakeholders for over 30 weeks without interruption.

The COVID-19 pandemic represents the most significant public health disaster since the 1918 influenza pandemic. During pandemics such as COVID-19, timely and reliable spatio-temporal forecasting of epidemic dynamics is crucial. Deep learning-based time series models for forecasting have recently gained popularity and have been successfully used for epidemic forecasting. Here we focus on the design and analysis of deep learning-based models for COVID-19 forecasting. We implement multiple recurrent neural network-based deep learning models and combine them using the stacking ensemble technique. In order to incorporate the effects of multiple factors in COVID-19 spread, we consider multiple sources such as COVID-19 confirmed and death case count data and testing data for better predictions. To overcome the sparsity of training data and to address the dynamic correlation of the disease, we propose clustering-based training for high-resolution forecasting. The methods help us to identify the similar trends of certain groups of regions due to various spatio-temporal effects. We examine the proposed method for forecasting weekly COVID-19 new confirmed cases at county-, state-, and country-level. A comprehensive comparison between different time series models in COVID-19 context is conducted and analyzed. The results show that simple deep learning models can achieve comparable or better performance when compared with more complicated models. We are currently integrating our methods as a part of our weekly forecasts that we provide state and federal authorities.

Deep learning classifiers are assisting humans in making decisions and hence the user's trust in these models is of paramount importance. Trust is often a function of constant behavior. From an AI model perspective it means given the same input the user would expect the same output, especially for correct outputs, or in other words consistently correct outputs. This paper studies a model behavior in the context of periodic retraining of deployed models where the outputs from successive generations of the models might not agree on the correct labels assigned to the same input. We formally define consistency and correct-consistency of a learning model. We prove that consistency and correct-consistency of an ensemble learner is not less than the average consistency and correct-consistency of individual learners and correct-consistency can be improved with a probability by combining learners with accuracy not less than the average accuracy of ensemble component learners. To validate the theory using three datasets and two state-of-the-art deep learning classifiers we also propose an efficient dynamic snapshot ensemble method and demonstrate its value. Code for our algorithm is available at https://github.com/christa60/dynens.

Influenza-like illness (ILI) is among the most common diseases worldwide. Producing timely, well-informed, and reliable forecasts for ILI is crucial for preparedness and optimal interventions. In this work, we focus on short-term but highresolution forecasting and propose DEFSI (Deep Learning Based Epidemic Forecasting with Synthetic Information), an epidemic forecasting framework that integrates the strengths of artificial neural networks and causal methods. In DEFSI, we build a two-branch neural network structure to take both within-season observations and between-season observations as features. The model is trained on geographically highresolution synthetic data. It enables detailed forecasting when high-resolution surveillance data is not available. Furthermore, the model is provided with better generalizability and physical consistency. Our method achieves comparable/better performance than state-of-the-art methods for short-term ILI forecasting at the state level. For high-resolution forecasting at the county level, DEFSI significantly outperforms the other methods.

Prophylactic interventions such as vaccine allocation are some of the most effective public health policy planning tools. The supply of vaccines, however, is limited and an important challenge is to optimally allocate the vaccines to minimize epidemic impact. This resource allocation question (which we refer to as VaccIntDesign) has multiple dimensions: when, where, to whom, etc. Most of the existing literature in this topic deals with the latter (to whom), proposing policies that prioritize individuals by age and disease risk. However, since seasonal influenza spread has a typical spatial trend, and due to the temporal constraints enforced by the availability schedule, the when and where problems become equally, if not more, relevant. In this paper, we study the VaccIntDesign problem in the context of seasonal influenza spread in the United States. We develop a national scale metapopulation model for influenza that integrates both short and long distance human mobility, along with realistic data on vaccine uptake. We also design GreedyAlloc, a greedy algorithm for allocating the vaccine supply at the state level under temporal constraints and show that such a strategy improves over the current baseline of pro-rata allocation, and the improvement is more pronounced for higher vaccine efficacy and moderate flu season intensity. Further, the resulting strategy resembles a ring vaccination applied spatiallyacross the US.

Linking multiple accounts owned by the same user across different online social networks (OSNs) is an important issue in social networks, known as identity reconciliation. Graph matching is one of popular techniques to solve this problem by identifying a map that matches a set of vertices across different OSNs. Among them, percolation-based graph matching (PGM) has been explored to identify entities belonging to a same user across two different networks based on a set of initial pre-matched seed nodes and graph structural information. However, existing PGM algorithms have been applied in only undirected networks while many OSNs are represented by directional relationships (e.g., followers or followees in Twitter or Facebook). For PGM to be applicable in real world OSNs represented by directed networks with a small set of overlapping vertices, we propose a percolation-based directed graph matching algorithm, namely PDGM, by considering the following two key features: (1) similarity of two nodes based on directional relationships (i.e., outgoing edges vs. incoming edges); and (2) celebrity penalty such as penalty given for nodes with a high in-degree. Through the extensive simulation experiments, our results show that the proposed PDGM outperforms the baseline PGM counterpart that does not consider either directional relationships or celebrity penalty.

Over the past decades, numerous techniques have been developed to forecast the temporal evolution of epidemic outbreaks. This paper proposes an approach that combines high resolution agent-based models using realistic social contact networks for simulating epidemic evolution with a particle filter based method for assimilation based forecasting. Agent-based modeling using realistic social contact networks provides two key advantages: (i) they capture the causal processes underlying the epidemic and hence are useful to understand the role of interventions on the course of the epidemics - typically time series models cannot capture this and as a result often do not perform well in such situations; (ii) they provide detailed forecast information - this allows us to produce forecast at high levels of temporal, spatial and social granularity. We also propose a new variation of particle filter technique called beam search particle filtering. The modification allows us to more efficiently search the parameter space which is necessitated by the fact that agent-based techniques are computationally expensive. We illustrate our methodology on the synthetic dataset of Ebola provided as a part of the NSF/NIH Ebola forecasting challenge. Our results show the efficacy of the proposed approach and suggest that agent-based causal models can be combined with filtering techniques to yield a new class of assimilation models for infectious disease forecasting.

Infectious disease epidemics such as influenza and Ebola pose a serious threat to global public health. It is crucial to characterize the disease and the evolution of the ongoing epidemic efficiently and accurately. Computational epidemiology can model the disease progress and underlying contact network, but suffers from the lack of real-time and fine-grained surveillance data. Social media, on the other hand, provides timely and detailed disease surveillance, but is insensible to the underlying contact network and disease model. This paper proposes a novel semi-supervised deep learning framework that integrates the strengths of computational epidemiology and social media mining techniques. Specifically, this framework learns the social media users' health states and intervention actions in real time, which are regularized by the underlying disease model and contact network. Conversely, the learned knowledge from social media can be fed into computational epidemic model to improve the efficiency and accuracy of disease diffusion modeling. We propose an online optimization algorithm to substantialize the above interactive learning process iteratively to achieve a consistent stage of the integration. The extensive experimental results demonstrated that our approach can effectively characterize the spatio-temporal disease diffusion, outperforming competing methods by a substantial margin on multiple metrics.

An edge switch is an operation on a network (graph) where two edges are selected randomly and one of their end vertices are swapped with each other. Usually, a sequence of these operations are performed to generate network perturbations having the same degree sequence of the original network. Edge switch operations have important applications in graph theory and network analysis, such as in generating random networks with a given degree sequence, modeling and analyzing dynamic networks (e.g., peer-to-peer networks), studying various dynamic phenomena over a network (e.g., disease dynamics over a social contact network). The growth of real-world networks motivates the need to develop efficient parallel algorithms for performing a large sequence of edge switch operations. The dependencies among successive edge switch operations and the requirement of keeping the graph simple (i.e., no self-loops or parallel edges) as the edges are switched lead to significant challenges in designing a parallel algorithm. Addressing these challenges requires complex synchronization and communication among the processors. In this paper, we present a distributed memory parallel algorithm for switching edges in massive networks (networks with billions of edges) and achieve a speedup factor of 85 with 1024 processors. One of the steps in our edge switch algorithm requires the computation of multinomial random variables in parallel. The paper presents the first non-trivial parallel algorithm for the problem. The algorithm achieves a speedup of 925 using 1024 processors.

We describe ISIS, a high-performance-computing-based application to support computational epidemiology of infectious diseases. ISIS has been developed over the last seven years in close coordination with public health and policy experts. It has been used in a number of important federal planning and response exercises. ISIS grew out of years of experience in developing and using HPC-oriented models of complex socially coupled systems. This identified the guiding principle that complex models will be used by domain experts only if they can do realistic analysis without becoming computing experts.

Modern epidemiological forecasts of common illnesses, such as the flu, rely on both traditional surveillance sources as well as digital surveillance data. However, most published studies have been retrospective. Concurrently, the reports about flu activity generally lags by several weeks and even when published are revised for several weeks more. We posit that effectively handling this uncertainty is one of the key challenges for a real-time prediction system in this sphere. In this paper, we present a detailed prospective analysis on the generation of robust quantitative predictions about temporal trends of flu activity, using several surrogate data sources for 15 Latin American countries. We present our findings about the limitations and possible advantages of correcting the uncertainty associated with official flu estimates. We also compare the prediction accuracy between model-level fusion of different surrogate data sources against data-level fusion. Finally, we present a novel matrix factorization approach using neighborhood embedding to predict flu case counts. Comparing our proposed ensemble method against several baseline methods helps us demarcate the importance of different data sources for the countries under consideration.

Modeling environments to study epidemic outbreaks can be used as decision support tools by decision makers to support public health policies. However, although the current high performance simulation engines have become adept at rapidly simulating disease diffusion, computational environments for exploring complex mitigation strategies are fairly rudimentary. Interactive simulations provide a natural way to study the complicated co-evolution of disease dynamics and public policies. In this paper, we formalize the problem of interactive simulations and present experimental results based on an interactive simulation platform that we have recently developed.

To respond to the serious threat of pandemics (e.g. 2009 H1N1 influenza) to human society, we developed Indemics (Interactive Epidemic Simulation), an interactive, data intensive, high performance modeling environment for realtime pandemic planning, situation assessment, and course of action analysis. Indemics was built upon a model of interactive data intensive scientific computation, supporting online interactions between users and simulations and enabling epidemic simulations over detailed social contact networks and realistic representations of complex public policies and intervention strategies.

Large scale realistic epidemic simulations have recently become an increasingly important application of high-performance computing. We propose a parallel algorithm, EpiFast, based on a novel interpretation of the stochastic disease propagation in a contact network. We implement it using a master-slave computation model which allows scalability on distributed memory systems.

A flow of a commodity is said to be confluent if at any node all the flow of the commodity leaves along a single edge. In this article, we study single-commodity confluent flow problems, where we need to route given node demands to a single destination using a confluent flow. Single- and multi-commodity confluent flows arise in a variety of application areas, most notably in networking; in fact, most flows in the Internet are (multi-commodity) confluent flows since Internet routing is destination based. We present near-tight approximation algorithms, hardness results, and existence theorems for minimizing congestion in single-commodity confluent flows. The maximum edge congestion of a single-commodity confluent flow occurs at one of the incoming edges of the destination. Therefore, finding a minimum-congestion confluent flow is equivalent to the following problem: given a directed graph G with k sinks and non-negative demands on all the nodes of G , determine a confluent flow that routes every node demand to some sink such that the maximum congestion at a sink is minimized. The main result of this article is a polynomial-time algorithm for determining a confluent flow with congestion at most 1 + ln( k ) in G , if G admits a splittable flow with congestion at most 1. We complement this result in two directions. First, we present a graph G that admits a splittable flow with congestion at most 1, yet no confluent flow with congestion smaller than H k , the k th harmonic number, thus establishing tight upper and lower bounds to within an additive constant less than 1. Second, we show that it is NP-hard to approximate the congestion of an optimal confluent flow to within a factor of (log 2 k )/2, thus resolving the polynomial-time approximability to within a multiplicative constant. We also consider a demand maximization version of the problem. We show that if G admits a splittable flow of congestion at most 1, then a variant of the congestion minimization algorithm yields a confluent flow in G with congestion at most 1 that satisfies 1/3 fraction of total demand. We show that the gap between confluent flows and splittable flows is much smaller, if the underlying graph is k -connected. In particular, we prove that k -connected graphs with k sinks admit confluent flows of congestion less than C + d max , where C is the congestion of the best splittable flow, and d max is the maximum demand of any node in G . The proof of this existence theorem is non-constructive and relies on topological techniques introduced by Lovász.

Using shortest paths, the Internet scales very poorly with respect to congestion [2]. Two main reasons for using shortest paths are dilation (or delay) and size of routing tables. As the Internet grows, the small size of routing tables is important for scaling, but it does not require shortest paths. As long as the paths are confluent, the routing table size is unchanged. In this paper we study the confluent capacity of the Internet. We use the preferential attachment model [5] for the Internet, and all-pair uniform demand for the traffic pattern. Our main theoretical result is that the confluent congestion1 is within a logarithmic factor of the optimal splittable congestion and can be achieved using a simple randomized and distributed scheme called Locally Independent Rounding Algorithm (LIRA). We reinforce this result experimentally by employing simulations to demonstrate that for almost all instances the confluent congestion is (nearly) equal to the splittable congestion. Thus we conclude that the Internet scales well using confluent paths. We combine known results on expanders and the expansion properties of the preferential attachment model to show that for almost all Internet-like networks, we can find a confluent flow that simultaneously achieves O(log n)- approximate congestion and O(1)-approximate dilation. We confirm, using simulations, the intuition that confluence does not come at the cost of dilation.

In this paper, we propose to address the energy efficient routing problem in multi-hop wireless networks with accumulative relay. In the accumulative relay model, partially overheard signals of previous transmissions for the same packet are used to decode it using a maximal ratio combiner technique [J.G. Proakis, 2001]. Therefore, additional energy saving can be achieved over traditional energy efficient routing. The idea of accumulative relay originates from the study of relay channel in information theory with a main focus on network capacity. It has been independently applied to minimum-energy broadcasting in L.G. Manish Agrawal et al. (2004), I. Maric and R. Yates (2002). We formulate the minimum energy accumulative routing problem (MEAR) and study it. We obtain hardness of approximation results counterbalanced with good heuristic solutions which we validate using simulations. Without energy accumulation, the classic shortest path (SP) algorithm finds the minimum energy path for a source-destination pair. However, we show that with energy accumulation, the SP can be arbitrarily bad. We turn our attention to heuristics and show that any optimal solution of MEAR can be converted to a canonical form - wave path. Armed with this insight, we develop a polynomial time heuristic to efficiently search over the space of all wavepaths. Simulation results show that our heuristic can provide more than 30% energy saving over minimum energy routing without accumulative relay. We also discuss the implementation issues of such a scheme.

A flow is said to be confluent if at any node all the flow leaves along a single edge. Given a directed graph G with k sinks and non-negative demands on all the nodes of G, we consider the problem of determining a confluent flow that routes every node demand to some sink such that the maximum congestion at a sink is minimized. Confluent flows arise in a variety of application areas, most notably in networking; in fact, most flows in the Internet are confluent since Internet routing is destination based.We present near-tight approximation algorithms, hardness results, and existence theorems for confluent flows. The main result of this paper is a polynomial-time algorithm for determining a confluent flow with congestion at most 1 + ln(k) in G, if G admits a splittable flow with congestion at most 1. We complement this result in two directions. First, we present a graph G that admits a splittable flow with congestion at most 1, yet no confluent flow with congestion smaller than Hk, thus establishing tight upper and lower bounds to within an additive constant less than 1. Second, we show that it is NP-hard to approximate the congestion of an optimal confluent flow to within a factor of (lg k)/2, thus resolving the polynomial-time approximability to within a multiplicative constant. We also consider a demand maximization version of the problem. We show that if G admits a splittable flow of congestion at most 1, then a variant of the congestion minimization algorithm yields a confluent flow in G with congestion at most 1 that satisfies 1/3 fraction of total demand.We show that the gap between confluent flows and splittable flows is much smaller, if the underlying graph were k connected. In particular, we prove that k-connected graphs with k sinks admit confluent flows of congestion less than C + dmax, where C is the congestion of the best splittable flow, and dmax is the maximum demand of any node in G. The proof of this existence theorem is non-constructive and relies on topological techniques introduced in [16].

In this paper we investigate the problem ofdetermining confluent flows with minimum congestion. A flow of a given commodity is said to be confluent if at any node all the flow of the commodity departs along a single edge. Confluent flows appear in a variety of application areas ranging from wireless communications to evacuations; in fact, most flows in the Internet are confluent since Internet routing is destination based.We consider the single commodity confluent flow problem, in which we are given an n-node directed network G, a sink t and supplies at each node, and the goal is to find a confluent flow that routes all the supplies to the sink while minimizing the maximum edge congestion. Our main result is an approximation algorithm, based on randomized rounding, for the special case when all the supplies are uniform; the algorithm finds a confluent flow with edge congestion O(C2 log3 n) where C is the node congestion of an optimal splittable flow. This implies an Õ(√n) approximation algorithm for the problem. Our result relies on the analysis of a natural probabilistic process defined on directed acyclic graphs, that may be of independent interest.For tree networks, we present an optimal polynomial-time algorithm for a multi-sink generalization of the above confluent flow problem. We show that it is NP-hard to approximate the congestion of the optimal confluent flow for general networks to within a factor of 4/3. We also establish a lower bound on the gap between confluent and splittable flows, and consider multicommodity and fractional versions of confluent flow problems.