1.0 Overview

The objective of the Network Sciences Division is to discover mathematical principles to describe ever present networks in all walks of life (e.g., organic, social, electronic) and, in particular, the emergent properties of networks. Study of Network Science is necessarily multi-disciplinary drawing on tools and techniques from statistical mechanics, information theory, computer science, control theory and social sciences to studies interactions at large scale, be they in swarms of insects or ant colonies, human societies, or networked autonomous systems. The basic principles discovered should, in turn, lead to creation of algorithms and autonomous systems that can be used to reason across data generated from disparate sources, including, sensor networks, wireless networks, and adversarial human networks. Research in this Division has applications to a wide variety of developmental efforts and contributes to the solution of technology-related problems throughout the Army's Future Force operational goals engendered by principles such as the Net-Centric Warfare and Third Offset strategy.

Research in the Network Science Division, while primarily driven by discovery of foundational principles, will, however, be cognizant of and contribute to providing crucial underpinning support to ARL's Information Sciences Campaign. In particular, the following goals within ARL's Information Sciences Campaign are explicitly addressed including: (i) assessment and control of behavior goal by creating new methods in design and controllability of composite and multi-genre networks; (ii) social effects and human-machine interaction through the exploration of social and cognitive networks, and generation of intelligent actions in a mix of information agents and humans; (iii) unconventional communication networks and adaptive by making information available at the tactical edge while taking limited bandwidth and human information interaction modalities into account, and (iv) taming flash-flood of information available at the tactical edge.

The Network Science Division hosts four main programs in Wireless and Hybrid Communication Networks, Social and Cognitive Networks, Intelligent Information Networks, Multi-Agent Network Control, and an international research program in Network Science and Intelligent Systems. The boundary between these programs is fluid and, thus, a research topic might fall in more than one program area. Furthermore, there are shared interests between the Social Sciences program in Physical Sciences Directorate and Social and Cognitive Networks (SCN) program in Network Science, with SCN paying special attention to the human dimension from a network science perspective, including study of connections between interdependent people (such as teams), between social systems and cognitive processing (such as collective learning and decision making), and between humans and machines, using tools and techniques from computer science to further the study of social and cognitive processes of humans embedded in large social, interconnected systems. It is perhaps worth emphasizing that research which elucidates and defines the common underpinning science cutting across different types of networks is particularly of interest to Network Science Division.

The Division's research areas are currently focused on five research areas that include one program focused on international research. The titles, scopes and points of contact for these programs, each of which address general aspects of basic research in network science, are listed in the following subsections. A small number of symposia, conferences and workshops are also supported in part or in whole to provide an exchange of ideas in areas of Army interest.

1.1 Wireless and Hybrid Communication Networks

This program is concerned with the application of emerging network science to wireless communication networks and human social networks. In both of these areas, DoD and particularly Army unique problems will be investigated. For communications networks, tactical multi-hop wireless communications which lack fixed infrastructure will be investigated. Also of interest is the interaction of the communications network with social networks and information networks.

1.1.1 Wireless Network Theory

Research is required in the broad area of wireless network science including fundamental limits, performance characterization, novel architectures, and high fidelity simulation. Metrics, fundamental limits, and performance need to be characterized for multi-hop wireless networks with mobility, node loss, and bursty traffic. New simulation techniques are necessary to allow for very large simulations without losing the fidelity at the physical layer that is necessary for realistic results.

The new communications network paradigms of Software Defined Networking (SDN) and Network Function Virtualization (NFV) is also of interest as applied to wireless and hybrid networks. Benefits of SDN/NFV include performance improvement, reduced predeployment setup, network protocol agility during the mission, and facilitating differential quality of service. Although not directly applicable to wireless multi-hop networks, concepts from SDN/NVF could be adapted for wireless networks. Control could not be at a single vulnerable node, but concepts of redundancy, and hierarchical control could be utilized. Overhead of in-band signaling would have to be taken into account, since a separate channel for out-of-band signal will likely not be available.

1.1.2 Mobile Ad Hoc and Sensor Networks

Networks serving Army needs must operate in highly dynamic environments with limited infrastructure support. Nodes in such networks often have only noisy local information and must operate in a decentralized fashion. New research approaches are needed to explicitly account for the lack of full network state information and the cost of coordination.

Adaptive networking techniques are needed for the dynamic allocation of resources based on operation needs, traffic characteristics, dynamic topologies, interference conditions, and security considerations. Networks need to be able to adapt to varied quality of information (QoI) requirements of various applications. Dynamic optimization and learning methods are needed to discover and capture communication and networking opportunities. Networking and sensing architectures for cognitive mobile ad hoc networks needs to be developed with qualitative and quantitative performance measures, and the impacts of mobility, fading, and multi-user interference needs to be investigated.

Networking in combat operations needs to cope with the presence of passive or active adversaries of various types. New signal processing, information theory, and networking theory methodologies are needed to provide reliable and efficient communications in the presence of various adversarial actions. There is a need to characterize fundamental tradeoffs among various conflicting objectives: between secrecy/anonymity vs. rate of communications vs. resource consumption, the need for rapid response and the requirement of authentication and security.

Providing energy efficient sensor networking under different operating conditions presents difficult technical challenges. New techniques are needed for sensor fusion using Shannon and non-Shannon information theoretic metrics for bandwidth and energy efficiency. Distributed synchronization methods with limited information exchange for varying network topology and heterogeneous delay requirements are needed. Research should investigate fundamental performance bounds and characterize the tradeoffs between conflicting metrics. Concepts from emerging Internet of Things (IoT) could be of interest in investigating the Battlefield IoT.

1.1.3 Interactions Between Social and Communications Networks

The synergy among social networking and communication networking, particularly in a tactical mobile ad-hoc scenario, is a fertile research area that can provide significant advances for the design of new communication. Important questions include: How do human networks behave in the presence of constraints imposed by the communication network? How do we design communication networks to achieve a global mission among the human actors which may be evolving in time as more information gets revealed and circumstances change? Interaction between human and communications networks needs to be analyzed, such as the interaction of Quality of Service (QoS) communications goals with the requirements of the human network in a tactical scenario. There are many social networking aspects that are common to mobile ad-hoc networking needs such as distributed decision making, robustness, cooperation, self-organization, cluster formation, search and exploration, to name a few. Social Networking Analysis concepts have been recently used in routing and storing of information for Disruption (or Delay) Tolerant Networks (DTN), with some encouraging results.

Technical Point of Contact: Dr. Derya Cansever, email: derya.h.cansever.civ@mail.mil, (919) 549-4330

1.2 Social and Cognitive Networks

The goal of the Social and Cognitive Networks program is to understand human behaviors and cognitive processes leading to collective level phenomena particularly relevant in military settings with an emphasis on high performance teams and computational social science. Social networks are the underlying structure of interaction and exchanges between humans within both strategically designed and emergent or self-organized systems. Social networks allow for collective actions in which groups of people can communicate, collaborate, organize, mobilize, attack, and defend. The changing nature of DoD's missions greatly increase the need for models that capture the cognitive, organizational, and cultural factors that drive activities of co-present, virtual, or distributed groups, teams, and populations. Better understanding the human dimension of complexity will provide critical insights about emerging phenomena, social diffusion and propagation, thresholds, and tipping points.

The Social and Cognitive Networks program supports projects that contribute substantive knowledge to theories about human behavior and interaction and make methodological advancements in modeling and analyzing social network structures. The U.S. Army is particularly interested in research that uses defense-relevant empirical data to feed into computational and mathematical models of human interaction. As such, this program funds projects successful in blending theories and methods from the social sciences with rigorous computational methods from computer science and mathematical modeling. Advances in this program are expected to lead to development of measures, theories, and models that capture behavioral and cognitive processes leading to emergent phenomena in teams, organizations, and populations.

1.2.1 Human Behavior and Interaction

This program supports research from disciplines such as communication, health and behavioral science, I/O and social psychology, library and information science, management science, and sociology that use a social networks lens to focus on the ways people think and interact whereby creating higher-order systems. Topics of interest include social influence, leadership, trust, team science, cooperation and competition, and crisis management. Such social influence and opinion dynamics research could focus on the formation and dissolution of civic-minded and violent ideological networks, mobilization of benign to hostile political movements, propagation of and enduring changes in attitudes leading to populations reaching consensus or contested states, and network-based interventions. Furthermore, topics of particular interest include social effects of human-machine teaming; multi-team systems and multilevel (nested) systems; and health topics related to education, healthcare behaviors, disease propagation, and wellness from a social networks perspective.

1.2.2 Information and Knowledge Management

This program supports social network centric research to study the ways people learn individually and collectively and how they utilize that information for decision making and goal attainment. Examples of relevant topics include transactive memories, public goods, collective action, information sharing, information fidelity, diffusion and propagation dynamics, and collective decision-making. Diffusion dynamics research will develop mechanistic understanding of opinion and behavior change associated with influence, contagion, and other social propagation processes. Collective decision-making research will contribute fundamental theories and models to predict, evaluate and simulate how teams organize, exchange information, build knowledge, influence, adapt, learn, and build consensus using cooperative strategies and emergent capabilities.

1.2.3 Social Network Analysis

In addition to the topical areas identified above, this program supports methodological advancements for social network analysis. Methodological research in this program will focus on important advances in exponential random graph models (ERGMs or p*), object oriented agent-based models, computational models, and dynamic simulations that resolve network modeling issues. Such research will focus on scalability of networks, hierarchical or multilevel (nested) systems, longitudinal networks, social influence models, network resilience, techniques to deal with missing, incomplete, or inaccurate network data, and techniques to deal with visualizing multilevel multimodal networks. Scalability and dimensionality research will identify overarching mechanisms that span scales and dimensions of human systems that will parameterize, model, and predict both small group and big data network models. Data accuracy research focuses on investigating effects of measurement error on metrics and inferences due to incomplete (missing or inaccurate) network data. These projects could include research examining small group dynamics within big data sets; multi-level models that account for nested cognitive, social, cultural, physical dimensions of systems; or link and subgroup estimation algorithms to deal with incomplete data and clandestine activities.

Technical Point of Contact: Dr. Edward T. Palazzolo, email: edward.t.palazzolo.civ@mail.mil, (919) 549-4234

1.3 Intelligent Information Networks

The overall objective of the Intelligent Information Networks program is to augment human decision makers (both commanders and soldiers) with enhanced-embedded battlefield intelligence that will provide them with the necessary situational awareness, reconnaissance, and decision making tools to decisively defeat any future adversarial threats which is in line with the DoD's adoption of net-centric warfare, variously defined as flattening the information space to interconnect soldiers and commanders to provide instantaneous access to information, knowledge, and situational awareness. Given this goal, it becomes necessary to understand (a) fundamentals of what intelligence means in the context of autonomous systems and how to build intelligent systems especially as it relates to interaction amongst a network of humans and machines, and (b) foundational algorithmic issues in representation and reasoning about networks inherent in societies and nature.

1.3.1 Integrated Intelligence, Theory of Mind, and Collective Intelligence

Intelligence emanates from several components acting in synergistic ways, be it human cognition that embodies cumulative effects of various separate components of the human brain or the collective intelligence of a network of agents (humans, birds, insects, etc.). Fundamental questions that need to be answered include: What is the least amount of knowledge necessary to bootstrap learning in autonomous systems (or in a network)? How can joint reasoning over various components be carried out (vision, knowledge representation, reasoning, planning, for instance) to obtain a sum that is more than its parts? Are there viable, computable theories of mind that can realistically implement reflection and meta-cognition? Can wisdom of crowds be harnessed to solve problems of importance to societies and problems that are deemed computationally hard? In particular, can a man in the middle of a man machine ensemble, or, more appropriately, can a crowd and machine ensemble solve inherently hard problems? What exactly are the limitations of wisdom of crowd, when the crowd is made of non-experts? Is there are a way for problems to be broken up so that a team of humans and machines can solve them together? Are there approaches to mechanism design that teases out intelligence inherent in a crowd (or society) of interest in this thrust? These are some of the questions whose solution would likely address basic research problems that are of interest to the program.

1.3.2 Information Networks

In order to model network effects it is necessary to algorithmically represent large networks and reason about them. Unfortunately, information about networks is seldom complete – data available might be missing crucial pieces of information, might have contradictory pieces of information, or could be approximate (with associated notions of uncertainty). Representing and reasoning about these networks requires advances in knowledge representation, graph and data mining, natural language processing, algorithmic graph theory, machine learning, and uncertainty quantification and reasoning. Examples include the emerging area of Graphons which provide new tools for generating and reasoning about graphs that occur in practice (satisfying power law distributions), but also provide new tools for Machine Learning. In particular, a major goal of this thrust are tools and techniques that allow data driven approaches to capturing latent relationships with powers to both explain and predict. Advances in this thrust would not only lead to improved autonomous systems and algorithms, but also enhanced-embedded battlefield intelligence with tools for creating necessary situational awareness, reconnaissance, and decision making. Finally, it should be noted that algorithmic notions of approximations, tight performance bounds, probabilistic guarantees, etc., would be major concerns of the solution space. Large graphs and voluminous data characterize problems in Network Science.

1.3.3 Adversarial Reasoning

Development of appropriate mathematical tools to model and reason about societies and cultures, that brings together tools from Game Theory, Social Sciences and Knowledge Representation. Research of interest includes, but is not limited to, Game Theory for security applications while accounting for bounded rationality, development of Game Theory based on data regarding cultural and adversarial groups, and Behavioral Game Theory that can explain intelligence in groups and societies.

Technical Point of Contact: Dr. Purush Iyer, e-mail: s.p.iyer.civ@mail.mil, (919) 549-4204

1.4 Multi-Agent Network Control

The objective of the Multi-Agent Network Control program is to establish the physical, mathematical and information processing foundations for the control of complex networks. The research program is concerned with developing novel mathematical abstractions and methods for the modeling and control of both individual agents as well as the collective behavior of large scale networks of heterogeneous multi-agent systems. In this regard, the term "agent" can span the biological, physical, and information and communication domain. Autonomy is central to program efforts as anticipated dynamics of the future battle space will require a greatly increased level of autonomy to enable the necessary mobility, sensor coverage, information flow, and responsiveness to support the military goals of information superiority, dominant maneuver, and precision engagement.

1.4.1 Distributed Control for Complex Networks

Distributed control techniques have played a major role in the study of networked systems. For example, they have been successfully used in robotics for replicating self-organized behaviors found in nature (e.g. bird flocking, fish schooling, and synchronization) and in developing applications such as formation control, rendezvous, robot coordination, and distributed estimation. A fundamental concept underlying these techniques is the notion of consensus. However, many control problems in complex networks cannot be framed as consensus problems. Hence, there is need for developing a new generation of distributed control methods for achieving more sophisticated control goals that are not amenable to a consensus-based formulation. In large-scale networks of interdependent dynamical systems this is likely to require hybrid control architectures that combine top-down and bottom-up design methods formally grounded in graph theory, dynamical systems theory, game theory, computational homology, and topology, amongst other disciplines. Research in this area should contribute to a better understanding of the tradeoffs between what can be achieved by a multi-agent system (e.g. controllability) versus (i) information processing requirements, network topology and computational overhead (ii) individual-agent control actuation capabilities, and (ii) degrees of autonomy and cognitive-behavioral issues arising from human-system interaction.

1.4.2 Analysis of Complex Co-Evolving Networks

The high-dimensionality and complex, evolving topologies in complex networks of heterogeneous agents with possibly asymmetric interactions requires new mathematical techniques for characterizing out-of-equilibrium dynamics and the likely equilibrium outcomes that may emerge over different time spans. Due to random perturbations, transitions between equilibria are inevitable with some transitions occurring more readily than others. Thus for a complete understanding of the dynamics of interacting agents, one must understand how and when equilibrium is likely to unravel, and which new equilibrium, if any, is likely to arise in its place.

1.4.3 Information Structure, Causality, and Dynamics for Control

Understanding information processing within and for controlling complex networks requires new tools for causality and topological inference. Major programmatic interests center on formulating abstractions of cognitive processes and perception for fundamentally changing the way information is exploited to enable control and re-shaping the actions of multiple agents and complex networks. Further questions include: How does control as an objective effect communication and information processing in complex networks? What are the fundamental limits of causality? How do we infer causality and perform estimation in complex networks.

Technical Point of Contact: Dr. Alfredo Garcia, e-mail: alfredo.a.garcia31.civ@mail.mil, (919) 549-4282

1.5 Network Science and Intelligent Systems (International Program)

As one of the ARO International Programs and part of the ARO Network Science Division portfolio, the Network Science and Intelligent Systems program is focused on supporting multidisciplinary research at institutions outside of the U.S., with the goal to accelerate new discoveries in network science and intelligent systems. Potential investigators should contact the Program's Technical Point of Contact for any questions regarding the geographic regions that can be considered for research proposals in this area.

1.5.1 Wireless Communications and Information Networks

Networks serving Army needs must operate in highly dynamic environments with limited infrastructure support. Networks will be disconnected, intermittent, and limited (DIL) environment, with limited state information and dynamic network connectivity and intermittent link connectivity as well as dynamic traffic load with various QoS constraints and priorities. Metrics, fundamental limits, and performance need to be characterized for tactical networks as well as new theory that will lead to reliable and efficient communications that meets QoS constraints. New algorithms and protocols that are more robust in the presence of various adversarial attacks are required.

One specific area of interest is software defined networking (SDN) that can operate within tactical DIL networks. Standard SDN are centralized utilizing a reliable control plane, which are not available in this environment. Research is needed to investigate if SDN can be used in this environment and how architectures and control algorithms need to be modified to meet QoS requirements.

1.5.2 Social Network Analysis and Visualization

Mathematical models for dynamics of large social networks are of interest. Modeling of dynamics of the social network are of interest to include both (and possibly co-evolving) structure and content (e.g., opinion dynamics). Hierarchical, multi-level, and composite network models are of interest to model interactions between networks. Scalability issues should be investigated to understand when they are applicable and models should be vetted against established sociological concepts and, if possible, using experimental data. Multi-scale visualization of social network data is important for presenting results of analysis to users and assisting in manual analysis. Missing, incomplete, and inaccurate data are issues that should be considered in network analysis techniques.

1.5.3 Dynamics of Interdependent and Multilayer Networks

In the emerging field of networks science, the importance of research into interdependent and multilayer networks, such as communications, social, and infrastructure networks, is becoming evident. The mathematics to understanding how to model, predict, and control such multi-layer and interdependent networks is an important research area.

1.5.4 Intelligent Systems

This part of the BAA invites proposals in the area of Network Inference, Data Mining, and Algorithmic Game Theory that supports aspects of Network Science, including understanding large groups, especially adversarial non-state actors. Work that advances Network Science by bringing new techniques from Game Theory, Machine Learning, Graph Algorithms, Reasoning under uncertainty, are welcome.

Technical Point of Contact: Dr. Robert Ulman, e-mail: robert.j.ulman.civ@mail.mil, +44-1895-626518


Last Update / Reviewed: August 15, 2017