Samarth Swarup

Research directions for service-oriented multiagent systems

Todays service-oriented systems realize many ideas from the research conducted a decade or so ago in multiagent systems. Because these two fields are so deeply connected, further advances in multiagent systems could feed into tomorrows successful service-oriented computing approaches. This article describes a 15-year roadmap for service-oriented multiagent system research.

A Study of Information Diffusion over a Realistic Social Network Model

Sociological models of human behavior can explain population-level phenomena within social systems; computer modeling can simulate a wide variety of scenarios and allow one to pose and test hypotheses about the social system. In this paper, we model and examine the spread of information through personal conversations in a simulated socio-technical network that provides a high degree of realism and a great deal of captured detail. To our knowledge thisis the first time information spread via conversation has been modeled against a statistically accurate simulation of peoples daily interactions within a specific urban or rural environment, capturing the points in time and space at which two people could converse, and providing a realistic basis formodeling human behavior during face-to-face interaction.We use a probabilistic model to decide whether two people will converse about a particular topic based on their similarity and familiarity. Similarity is modeled by matching selected demographic characteristics, while familiarity is modeled by the amount of contact required to convey information. We report our findings on the effects of familiarity and similarity on the spread of information over the social network. We resolve the results by age group, daily activities, time, household income, household size and examine the relative effect of these factors.For informal topics where little familiarity is required, shopping and recreational activities predominate; otherwise, home, work, and school predominate. We find that youths play a significant role in spreading information through a community rapidly, mainly through interactions in schools and recreational activities.

Blocking Simple and Complex Contagion by Edge Removal

Eliminating interactions among individuals is an important means of blocking contagion spread, e.g., closing schools during an epidemic or shutting down electronic communication channels during social unrest. We study contagion blocking in networked populations by identifying edges to remove from a network, thus blocking contagion transmission pathways. We formulate various problems to minimize contagion spread and show that some are efficiently solvable while others are formally hard. We also compare our hardness results to those from node blocking problems and show interesting differences between the two. Our main problem is not only hard, but also has no approximation guarantee, unless P=NP. Therefore, we devise a heuristic for the problem and compare its performance to state-of-the-art heuristics from the literature. We show, through results of 12 (network, heuristic) combinations on three real social networks, that our method offers considerable improvement in the ability to block contagions in weighted and unweighted networks. We also conduct a parametric study to understand the limitations of our approach.

Modeling the interaction between emergency communications and behavior in the aftermath of a disaster

We describe results from a computer simulation-based study of a large-scale, human-initiated crisis in a densely populated urban setting. We focus on the interaction between human behavior and the communication infrastructure in the aftermath of the crisis. We study the effects of sending emergency broadcasts immediately after the event, advising people to shelter in place, and show that this relatively mild intervention can have a large beneficial impact.

Semantic network analysis of vaccine sentiment in online social media

OBJECTIVE To examine current vaccine sentiment on social media by constructing and analyzing semantic networks of vaccine information from highly shared websites of Twitter users in the United States; and to assist public health communication of vaccines. BACKGROUND Vaccine hesitancy continues to contribute to suboptimal vaccination coverage in the United States, posing significant risk of disease outbreaks, yet remains poorly understood. METHODS We constructed semantic networks of vaccine information from internet articles shared by Twitter users in the United States. We analyzed resulting network topology, compared semantic differences, and identified the most salient concepts within networks expressing positive, negative, and neutral vaccine sentiment. RESULTS The semantic network of positive vaccine sentiment demonstrated greater cohesiveness in discourse compared to the larger, less-connected network of negative vaccine sentiment. The positive sentiment network centered around parents and focused on communicating health risks and benefits, highlighting medical concepts such as measles, autism, HPV vaccine, vaccine-autism link, meningococcal disease, and MMR vaccine. In contrast, the negative network centered around children and focused on organizational bodies such as CDC, vaccine industry, doctors, mainstream media, pharmaceutical companies, and United States. The prevalence of negative vaccine sentiment was demonstrated through diverse messaging, framed around skepticism and distrust of government organizations that communicate scientific evidence supporting positive vaccine benefits. CONCLUSION Semantic network analysis of vaccine sentiment in online social media can enhance understanding of the scope and variability of current attitudes and beliefs toward vaccines. Our study synthesizes quantitative and qualitative evidence from an interdisciplinary approach to better understand complex drivers of vaccine hesitancy for public health communication, to improve vaccine confidence and vaccination coverage in the United States.

Modeling the effect of transient populations on epidemics in Washington DC

Large numbers of transients visit big cities, where they come into contact with many people at crowded areas. However, epidemiological studies have not paid much attention to the role of this subpopulation in disease spread. We evaluate the effect of transients on epidemics by extending a synthetic population model for the Washington DC metro area to include leisure and business travelers. A synthetic population is obtained by combining multiple data sources to build a detailed minute-by-minute simulation of population interaction resulting in a contact network. We simulate an influenza-like illness over the contact network to evaluate the effects of transients on the number of infected residents. We find that there are significantly more infections when transients are considered. Since much population mixing happens at major tourism locations, we evaluate two targeted interventions: closing museums and promoting healthy behavior (such as the use of hand sanitizers, covering coughs, etc.) at museums. Surprisingly, closing museums has no beneficial effect. However, promoting healthy behavior at the museums can both reduce and delay the epidemic peak. We analytically derive the reproductive number and perform stability analysis using an ODE-based model.

The classification game: combining supervised learning and language evolution

We study the emergence of shared representations in a population of agents engaged in a supervised classification task, using a model called the classification game. We connect languages with tasks by treating the agents’ classification hypothesis space as an information channel. We show that by learning through the classification game, agents can implicitly perform complexity regularisation, which improves generalisation. Improved generalisation also means that the languages that emerge are well adapted to the given task. The improved language-task fit springs from the interplay of two opposing forces: the dynamics of collective learning impose a preference for simple representations, while the intricacy of the classification task imposes a pressure towards representations that are more complex. The push–pull of these two forces results in the emergence of a shared representation that is simple but not too simple. Our agents use artificial neural networks to solve the classification tasks they face, and a simple counting algorithm to learn a language as a form-meaning mapping. We present several experiments to demonstrate that both compositional and holistic languages can emerge in our system. We also demonstrate that the agents avoid overfitting on noisy data, and can learn some very difficult tasks through interaction, which they are unable to learn individually. Further, when the agents use simple recurrent networks to solve temporal classification tasks, we see the emergence of a rudimentary grammar, which does not have to be explicitly learned.

Computational Epidemiology in a Connected World

New technologies help epidemiologists model the socioeconomic context of disease outbreaks. Epidemiologists and computer scientists are developing new data-driven, high-performance-computing-powered inference engines to model the socioeconomic context and strategies necessary to counter disease outbreaks.

A general-purpose graph dynamical system modeling framework

We describe InterSim, a general purpose flexible framework for simulating graph dynamical systems (GDS) and their generalizations. GDS provide a powerful formalism to model and analyze agent-based systems (ABS) because there is a direct mapping between nodes and edges (which denote interactions) in a GDS and agents and interactions in an ABS, thereby providing InterSim with great expressive power. We describe the design, implementation, capabilities, and features of InterSim; e.g., it enables users to quickly produce simulations of ABS in many application domains. We present illustrative case studies that focus on the simulation of social phenomena. InterSim has been used to simulate networks with 4 million agents and to execute large parametric simulation studies.

Planning and response in the aftermath of a large crisis: an agent-based informatics framework

We present a synthetic information and modeling environment that can allow policy makers to study various counter-factual experiments in the event of a large human-initiated crisis. The specific scenario we consider is a ground detonation caused by an improvised nuclear device in a large urban region. In contrast to earlier work in this area that focuses largely on the prompt effects on human health and injury, we focus on co-evolution of individual and collective behavior and its interaction with the differentially damaged infrastructure. This allows us to study short term secondary and tertiary effects. The present environment is suitable for studying the dynamical outcomes over a two week period after the initial blast. A novel computing and data processing architecture is described; the architecture allows us to represent multiple co-evolving infrastructures and social networks at a highly resolved temporal, spatial, and individual scale. The representation allows us to study the emergent behavior of individuals as well as specific strategies to reduce casualties and injuries that exploit the spatial and temporal nature of the secondary and tertiary effects. A number of important conclusions are obtained using the modeling environment. For example, the studies decisively show that deploying ad hoc communication networks to reach individuals in the affected area is likely to have a significant impact on the overall casualties and injuries.