4
Information Sciences
The Panel on Information Science at the Army Research Laboratory (ARL) was charged with reviewing ARL research in the broad areas of information sciences—network sciences, cyber sciences, sensing and effecting (S&E), system intelligence and intelligent systems (SIIS), human-information interaction (HII), and atmospheric sciences. A 2-year cycle of review has been adopted for this purpose, with the focus in 2017 on reviewing activities in SE, SIIS, HII, and atmospheric sciences, and the focus in 2018 on reviewing research related to networks and communication and cybersecurity: detection and agility.
ARL research in the Information Sciences Campaign is focused on developing and enhancing science and technology (S&T) capabilities that allow for the timely acquisition and use of high-quality information and knowledge at the tactical edge, for both strategic operations planning and mission deployment. Included in this approach are technological advances that support information acquisition, reasoning with such information, and support for decision-making activities, including collaborative communications. The research is expected to generate new technology to both manage and effectively use information flows in the battlespace of the future. Research in these areas falling under the broad categories of SE, SIIS, HII, and atmospheric sciences were reviewed in Adelphi, Maryland, from June 7-9, 2017. This chapter provides an evaluation of that work. Research falling under the broad areas of networks and communications and cybersecurity detection and agility was reviewed in the current cycle. The panel charged with the review in each of these subject areas was convened in Adelphi, Maryland, on May 22-24, 2018.
Research in sensing and effecting is organized around understanding of sensing capabilities and exploiting information gained through sensing to drive effectors. Research in this area focuses on understanding the sensed phenomena, including the effects of the environment on the phenomena. Both sensing and effecting necessitate detailed understanding of corresponding physical behaviors that generate and utilize data, an ability to extract relevant information from sensors of different modalities, an understanding of the complex and variable environments in which these sensors operate, and an ability to combine
information from multiple modalities and other contextual and environmental information. Research in effecting focuses on the understanding of transforming information into physical events, and provides insights into the nature and characteristics of related phenomenology, thereby helping design effectors that provide the desired physical effects. This work includes development of advanced algorithms to improve signal-to-noise ratios for all sensing modalities. The overarching goal of sensing and effecting is to deliver relevant information to decision makers in a timely manner to facilitate effective action.
System intelligence and intelligent systems research at ARL seeks to both understand and exploit interactions between information and intelligent systems, such as software agents or robots. The research portfolio includes a range of topics, from core machine learning, vision, and natural language organization and understanding to integration of information and decision making. The primary essential research areas (ERAs) reviewed include artificial intelligence and machine learning (AI and ML) and human-agent teaming (HAT). An important consideration in this work is to think of information as data in context, and to use this data to develop automated intelligence in perception, reasoning, planning, and collaborative decision making, with applications in cyber virtual environments or decision support systems. Research in this area complements efforts being undertaken in the Sciences for Maneuver Campaign involving intelligent systems concepts applied to vehicles or robotic platforms.
The Army faces significant and growing challenges in the area of human-information interaction. This is an extremely broad area with obvious linkages to human sciences and other information sciences programs. The growing use of unmanned vehicles and robots and the expanding influence of social media in planning and operations are creating a challenging communications environment for the Army of 2040. The need for effective communications between humans and machines, often without the ability to determine whether the communication is from and to a human or a software agent, introduces additional degrees of complexity. The need to operate effectively in this environment is creating a need for a new transdisciplinary science of HII. This is a crucial area for ARL basic research because many of the core issues are deeply intertwined. Mission requirements such as the need for diverse teams to operate under extreme time pressure in unpredictable, resource-constrained battlefield conditions are not a consideration in most academic research. This is an area in which ARL has a strong potential, and an imperative, to lead in research. Of course, as they do this, relevant academic work needs to also be monitored and exploited.
The characterization of the battlespace environment and prediction of optimal conditions for engaging an adversary with overwhelming force present significant challenges for the Army of the future. The Army is likely to find itself engaged in a number of challenging environments, from complex terrain to sprawling urban areas. There is a need, therefore, to characterize these diverse environments and develop accurate, relevant, and timely predictions of their future state on spatial and temporal scales useful to Army operations. The related need to collect and process accurate, relevant, and timely environmental characterizations in austere conditions and translate that data into actionable environmental intelligence for field commanders will also pose new challenges to the computational sciences community. The research being conducted by the Battlefield Environments Division is addressing these challenges, and includes a mix of analytical, computational, and experimental projects. There are clear linkages of the individual research projects to the nine ERAs considered essential to the success of Army operations in the 2030-2050 time frame.
The focus of ARL research in networks and communications is on studying the structure and dynamical behavior of networks, and on developing an understanding of the interactions between communications and both information and social networks. Of special interest in this work is developing the means to transfer information in a network efficiently, with a focus on algorithms and rules that govern the transfer of information, including the information format that promotes robustness and efficiency. There
is an emphasis on the development of unconventional alternative networks that embrace heterogeneous approaches to encoding and transmitting information, with a focus on robustness and survivability in harsh operating environments. ARL research is also directed at studying adaptive networks that emerge through interaction and co-evolution of different types of networks. The research seeks an improved understanding of adaptive protocols required for the control of such networks. Information delivery in a robust and reliable manner is an important objective, and the research portfolio includes development and validation of theories and models of storage, management, and transmission of information.
Research in the area of cybersecurity is focused on both understanding and exploiting the manner in which cyberattackers—both human and intelligent agents—interact with information. The defense of information systems and networks, robustness of information support systems for soldiers, and operations against adversarial systems are all within the scope of the research program. Understanding the cyber threat is an important focus of the work; this includes analysis and understanding of adversarial resources, learning about adversary tactics, and anticipating adversarial actions. The automated detection of hostile activities in cyberspace is another important consideration. Here, the focus of research resides in both algorithmic tools for detecting malicious activities as well as psychosocial and cognitive approaches to optimizing performance of human cyber analysts. A third area of research is on approaches to prevent hostile activities and to defeat such threats quickly and with minimal disruptive effects. Risk characterization is another important element of cybersecurity research, and focuses on metrics and algorithms for risk assessment and on risk data collection to support risk assessment.
SENSING AND EFFECTING
Accomplishments and Advancements
Research projects in sensing and effecting covered thematic areas of nonimaging sensors (acoustic, electric, magnetic, seismic), radar sensing and signal processing, image and video analytics, sensor and data fusion, and machine learning. The nonimaging research presentations focused on electric and magnetic field sensing and on acoustic classification. Radar sensing included advancements in landmine and improvised explosive device (IED) detection, adaptive radar transmit-waveform selection, and radar imaging in congested spectrum environments. Image and video understanding research included cross-modal face recognition, real-time image object recognition, and human action learning and recognition. Sensor fusion research included multimodal image fusion and understanding, combined text and video analytics, and multimodal fusion for detection and estimation. Machine learning research was used as a tool in several of the science and engineering (S&E) projects.
The research portfolio contains a mix of basic and applied research, and some of the projects show clear potential for transition into Army applications and products. Noteworthy programs include electric and magnetic field sensing, research on the next-generation IED and landmine detection platform, computational advances in electric field modeling, cross-modal face recognition, along with the development and dissemination of a cross-modal face recognition data set to the academic research community, and innovative approaches to fuse textual context with image features to improve machine learning of human activity.
Nonimaging Sensors
This area of research included projects in modeling of electric and magnetic fields, electric and magnetic field sensing technologies and exploitation algorithms, and acoustic classification algorithms.
The research on electric and magnetic field modeling represents high-quality algorithmic development, and seeks to apply computer science expertise to an electromagnetics problem. The research program contains a healthy mix of fundamental work related to understanding of phenomenology, sensor development and assessment, and algorithm and application advancements. The work reviewed was of high quality, and this group provides national leadership in this important area. Mentorship and development of early-career researchers is especially strong, and it is clear that ARL is nurturing the next generation of talent in this area.
Radar Sensing
Projects in this category included research in radar imaging and in adaptive transmit waveform selection in congested radio frequency interference (RFI) environments, and on further development of standoff landmine and IED detection. The research on radar imaging in RFI environments builds on a well-established program at ARL, and has potential to transition into operational use. The research on optimal transmit waveform selection in an RFI environment is in early stages, and is addressing an important problem that will enable radar sensing as the RF spectrum becomes increasingly crowded and complex. The research on a vehicle-mounted radar sensor for landmine and IED detection is advancing the capability in multisensor design and phenomenology, and is addressing important technical challenges needed for transition into operation.
Image and Video
Research projects in this grouping included cross-modal face recognition, real-time image object recognition, and human action learning and recognition. The cross-modal face recognition research is advancing the capability of facial recognition by comparing images collected in one modality with a catalogue of facial images from another modality. Researchers collected and disseminated a data set to the academic research community to facilitate development by others and grow collaborations with ARL. Research on real-time image object recognition is in the early stage, and seeks to develop algorithms that could be deployed on video sensors to automatically detect objects of interest from video streams. Research is also being conducted to develop algorithms to model and to learn human actions and activities from image sequences or video streams.
Sensor Fusion
These research projects included multimodal image fusion and understanding, combined text and video analytics, and multimodal fusion for detection and estimation. The project on multimodal image fusion and understanding applies algorithms to fuse point-cloud and other image data for use in scene situational awareness. Advancements in improving robustness in fusion of data from coarse-grained and fine-grained sensors, such as acoustics and imagery, were presented. Research attempting to improve human activity classification using text and linguistic information is in an early stage of development. The research on cross-modal face recognition is strong, and the data collection and dissemination efforts are commendable. The research on fusing point clouds with imagery was of good quality.
Machine Learning
This research included decentralized estimation and learning methods; in addition, machine learning was used as a tool to support several of the projects described in other areas discussed earlier. The distributed learning research was considered to be of high scientific quality. The demonstration was on a “toy” problem. It will be useful to assess the research as it matures to the point where it can be demonstrated on a relevant problem in an experimental test-bed.
Challenges and Opportunities
While the overall quality of the research was considered strong, it was not uniformly so across all presentations. Some researchers did not demonstrate good understanding of the existing state of the art. As a result, these presentations lacked a clear connection to existing research, and in some cases, it was not clear why existing techniques were inadequate to address the problem at hand. Researchers were generally able to articulate those aspects of their work that made it unique to Army needs and had Army relevance. However, this was not uniformly so, and it is advisable that all researchers develop a strong understanding of Army relevance and uniqueness of their work, as well as an ability to communicate this in an effective manner. Such understanding would contribute to developing research problem statements better aligned to needs, and consequently higher impact of successful research endeavors. Applied research programs would benefit from clear transition planning that includes mission fit to a present or an envisioned Army need. Basic research programs would benefit from at least some early consideration of this relevance.
The research staff at ARL is well qualified and equipped to take on problems of increased complexity and challenge. To this extent, the research problem statements in some instances were less ambitious than they could be. This was noted more often in cross-disciplinary endeavors. ARL has in its existing workforce, either within the S&T portfolio or in the larger ARL cadre, the research talent and expertise to formulate and pursue challenging research goals. Researchers pursuing such cross-disciplinary projects are encouraged to establish collaborations and mentoring relationships with experts in allied areas early in the problem formulation stage.
The posters, presentations, and oral summaries were generally good in providing sufficient details about assumptions and experiments to assess the quality of the work. Some presentations and posters lacked important detail about assumptions being made, and details about the experimental design setups that led to the reported performance results. It was in some cases difficult to understand what experiment was conducted and what significant observations or conclusions could be drawn from the results. Research impact and understanding would be improved by additional attention to these descriptions in the posters and oral presentations.
Nonimaging Sensors
The relevance of the work to the ARL key campaign initiatives (KCIs) and core campaign enablers (CCEs) was not very clear, and Army relevance and impact was hard to gauge. The acoustic classification research is considered as a collaborative project between the sensing and effecting and battlefield environment groups. Tighter collaboration and integration with the battlefield environment researchers would yield enhanced impact. The performance gains might be improved with a more integrated consideration of propagation physics in the sparsity models.
Radar Sensing
The research on radar algorithms in spectrum-congested environments would benefit from a stronger development of a transition strategy; this would help to focus the research to ensure addressing transition-relevant research issues. Additionally, with the rapidly shrinking radio frequency (RF) spectrum available to radar, the research would benefit by considering transmit waveform adaptation to effect power notching as an approach to avoid interference with communication bands. The cognitive radar research has opportunity for stronger impact if a more ambitious problem statement were adopted. In particular, more general waveforms that use more of the available spectrum could be considered. This research could consider sensing the available spectrum on both transmit and receive, and leveraging the cognitive radar research conducted by Defense Advanced Research Projects Agency (DARPA) and other Department of Defense (DoD) service laboratories.
Image and Video
The research in real-time video analytics is at an early stage, and the object recognition aspects were not completely clear. The research might benefit from a crisper and clear problem statement and methodology approach, and from a stronger, more integrated consideration of object recognition. In addition, a comparison to semisupervised learning using convolutional neural networks could improve the impact of the research. The research on human action recognition would benefit from a clearer presentation of the mathematical construct, experiment design, and experimental results.
Sensor Fusion
The research on cross-modal face recognition would benefit from consideration of larger subject database sizes, and also from the inclusion of battlefield environment atmospheric effects (such as dust or smoke) on recognition performance. As regards the research on fusing point clouds with imagery, other research groups, both within ARL and within the Night Vision Laboratory, are considering related problems, and researchers are encouraged to understand and leverage other similar research in this area. The research on improved multimodal fusion would benefit from a clearer articulation of the problem statement, and by framing the research approach more clearly in the context of existing research literature on the topic. It was difficult to assess how the proposed approach yielded the stated performance gains.
SYSTEM INTELLIGENCE AND INTELLIGENT SYSTEMS
Accomplishments and Advancements
SIIS research has produced key results in areas relevant to Army needs, including the understanding and analysis of complex environments and streaming data, navigation, exploration, and mapping of the physical world. The work on unsupervised learning of semantic labels in streaming data, and the synergies between visual analysis and efficient exploration of environments, is noteworthy. It has identified gaps in the state of the art and is being disseminated in leading venues.
The SIIS research portfolio also includes work related to text analysis, language understanding and dialogue, information integration, and decision making. The approach of collaborating with researchers throughout ARL as well as on the outside to develop a continuum of work, ranging from information analysis (in SIIS) to decision support (in HII), will yield good dividends. The research on integrating
event ontologies to increase the coverage of events provides near- to midterm utility. In the longer term, approaches that combine curated resources and more data-driven approaches as well as integrating text and video could provide added benefits. Research on natural language dialogue to support human-agent interaction, and agent intervention for decision and information integration, is important in supporting efficient communication and information analysis. Much of this work is focused on novel methods to collect data to support the design of such systems, and results were preliminary but promising.
The sample of research presentations and posters reviewed in this cycle are in broad subject areas of information understanding, information fusion, and computational intelligence.
Ontology for Event Understanding
A critical issue for analysts and decision makers is to rapidly and accurately identify events of concern in documents. An important component of such an information extraction data pipeline is an ontology, and the ongoing work explores representations that could be used to encode and integrate previous ontologies to obtain better coverage. The research seeks to evaluate how the integration affects information and decision tasks, and it is of good quality.
Knowledge Representation and Modeling for Understanding Information Amalgamation
This work seeks to use knowledge representation to define a computational model of the value of information (VoI). In previous research, the team developed a fuzzy logic-based approach to model a VoI metric across various information characteristics (e.g., reliability of source and credibility of content) and mission contexts. The current research extends this by investigating how analysts combine complementary and contradictory information. Some potentially interesting trends were identified, but the data are sparse, and the result may be context specific. The underlying effort to use cognitive modeling to assist in tactical intelligence data fusion has great midterm potential, and the linkage to experts in cognitive modeling is laudable.
Toward Natural Dialogue with Robots: BOT Language
This project seeks to advance the state of the art in natural language dialogue processing to support human-agent interaction in the context of a collaborative search-and-navigation setting. A novel “Wizard of Oz” methodology was developed to collect a corpus of naturally occurring variation in language in this setting. Initial results have helped in understanding of dialogue structures in these settings.
Air-Ground Robot Team Surveillance of Complex 3D Environments
The focus of this work is to improve autonomous surveillance via robots, with a human operator helping to focus attention on relevant areas. Starting from the assumption of a known map, the robot chooses a plan that is within a resource budget and maximizes coverage. The human operator can indicate one or more regions of interest, and the robot adjusts its plan to prioritize these areas. This effort has advanced beyond simulation, and the evaluation is being done with real robots in interesting environments. Future work on this project includes the inclusion of additional constraints, including the ability to select between ground or air vehicles.
Parsimonious Online Learning with Kernels
This high-quality work is focused on an important subset of learning techniques that are relevant to Army needs. In particular, the research is focused on the “5 D’s” (dynamic, dirty, deceptive, dinky data), and the approach is predicated on the use of online, parametric, kernel-based techniques. Results included an evaluation for the proposed approach based on simulations, as well as some work toward a target application predicting slip or oversteer based on images of the surface. The researcher was effective in providing context for the work and identifying limitations.
Knowledge Representation and Modeling for Understanding Information Amalgamation
The objective of this work was to use knowledge representation to define a computational model of the VoI. Previous research had focused on the development of a fuzzy logic-based approach to model a VoI metric across various information characteristics (e.g., reliability of source and credibility of content) and mission contexts. The current research extends this by investigating how analysts combine complementary and contradictory information. Some potentially interesting trends were identified, but the data are sparse, and the result may be context specific. The underlying effort to use cognitive modeling to assist in tactical intelligence data fusion has great midterm potential, and the linkage to experts in cognitive modeling is to be applauded and supported.
Reasoning under Uncertainty via Subjective Logic Bayesian Networks
Inference is a challenging task in the presence of noisy, sparse, and untrustworthy data. This work introduces subjective belief propagation, a new technique that extends belief propagation to efficiently infer uncertain marginal probabilities over subjective Bayesian networks. This is a promising technique to provide robust inference over uncertain relations.
Deductive and Analogical Reasoning on a Semantically Embedded Knowledge Graph
This research is focused on reasoning with brittle and incomplete information in knowledge bases and combines deductive and analogical reasoning and association by mapping information into a vector space, and by using the geometry of the space to find “related” paths. Previous work has looked at vector relations in embedded knowledge representations, but none has combined this with deductive reasoning. The ability to combine symbolic and continuous representations is a promising direction.
Unsupervised Semantic Scene Labeling for Streaming Data
The goal of this work is to augment visual perception by automatically extracting semantic groupings of objects found in video. This project was particularly strong, both for Army relevance and in its execution. In particular, the use of unsupervised techniques means that the approach does not require tedious labeling, a difficult task for video. Furthermore, the approach is applicable to long-running video without requiring excessive storage. To restrict the propagation of error introduced by unsupervised learning, the approach incorporates an ensemble gained from training the model on a sliding window of frames with periodic resets. The work compares favorably to previous techniques such as over- and undersegmentation entropy and graph-based techniques. Results of this research have been published in papers presented at top vision conferences.
Autonomous Mobile Information Collection Using Value of Information Enhanced Belief Approach
This research focuses on a framework for planning and optimization in multirobot environments. By using a model based on partially observable Markov decision processes, and by using the VoI to help prioritize, the approach is expected to pick better plans to simultaneously meet mission objectives and maximize information about the environment. The research is at an early stage of development, and a simple prototype simulation has been used in early investigations and described in a paper presented at this year’s International Conference on Robotics and Automation (ICRA), a top-tier conference in robotics.
A Crowd-Sourced Integration of Intervention Thresholds for Decision Agents
This research project seeks to develop an approach for supporting human decision making with automated agents. A new methodology was developed in which a single-player game was developed to help an analyst solve an information-based problem that was deployed on Amazon Mechanical Turk (AMT) with hundreds of participants. The primary outcome seems to be building experience using crowd-sourcing platforms for these experiments, and provides an initial indication that agents can help support these decision-making tasks.
Robot Demonstration
A robot capable of autonomous navigation, exploration, and mapping was demonstrated. The robot was tasked with mapping a complicated space via light detection and ranging (lidar), inertial, and odometry-based sensors. The on-board algorithm was designed to select an optimal pathway to map the interior, and was sufficiently capable to avoid dead-end paths that might impede its task. Additionally, a user could draw a polygon representing a region of interest, and the robot would dynamically adapt its plan to map the desired space. This demonstration was an effective illustration of several ARL technology thrusts integrated in a single robotic platform. It was nice to see an end-to-end demonstration of the Army technologies of high relevance working together. The laboratory space where the demonstration was staged represents an effective resource for researchers from all over ARL, as well as academic open campus partners. In particular, the support for reconfiguring the space, the ability to fly small unmanned aerial vehicles (UAVs), Global Positioning System (GPS) denial, and calibrated measurement equipment (e.g., the VICON camera system) are all features that make the space valuable for ongoing research.
Challenges and Opportunities
It is evident that artificial intelligence and machine learning have been assigned priority as a crucial area, and that ARL is organizing its research portfolio, particularly in SIIS, to address key gaps relevant to the Army. There is specific emphasis in areas including learning with sparse and noisy data, learning in adversarial settings, unsupervised learning, vision, decision making under uncertainty, and natural language processing. The research projects are appropriately resourced. Bright early-career researchers have been hired, and there appears to be a healthy cross-disciplinary integration. Overall, the facilities and computational resources seem adequate. The facility conducting experiments with robots, UAVs, and humans is especially impressive. This facility will help attract staff and visiting researchers to the laboratory. Engagements with universities seem to be a great source of collaborations and new talent. It will be
important to continue to invest in this area, both internally and especially with university and industrial partners, where the pace of development in this area is extremely rapid. ARL plans for establishing the Intelligent Systems Center (ISC) to facilitate open campus collaboration and unification of effort to take intelligent systems from tools to teammates are an indicator to grow and promote partnerships.
There is some concern about researchers being isolated from each other, as well as from the outside world. There was some evidence that SIIS researchers within ARL were not always aware of similar or parallel efforts. This is not unique to ARL, and all government laboratories generally can benefit from closer interaction with industry and universities around Army-relevant needs. One way to further enhance this awareness might be to create and share relevant data sets.
An opportunity exists for SIIS researchers to look for interesting science with applications of machine learning in domains where the Army may have a great amount of labeled data, and perhaps even structured data, such as logistics or medical records. There may be immediate gains in the application of methods that have been successfully used by industry (e.g., deep neural nets). Engaging with such applications will help keep SIIS researchers current on big data tools and methods developed elsewhere.
Ontology for Event Understanding
In addition to evaluating the impact of extended coverage on an end-to-end pipeline task, an important future direction would be to integrate curated ontologies with more data-driven approaches.
Toward Natural Dialogue with Robots: BOT Language
Moving from a controlled experimental setting to one with operational components would accelerate data collection and research progress. In addition, it might be interesting to consider the integration of language with richer multimodal signals (vision, mapping).
Reasoning under Uncertainty via Subjective Logic Bayesian Networks
Generalizing the work to realistic applications will require a number of advances, such as moving from binary to multinomial, inference over directed acyclic graphs (as opposed to trees), and general graphical models.
A Crowd-Sourced Integration of Intervention Thresholds for Decision Agents
The findings were preliminary and would benefit from a more detailed statistical analysis of the relationship between decision making and agent guidance, and a broader framing of the decision-making options. The work has been published in defense-based sensing venues, and might benefit from a broader AI or decision-making setting.
HUMAN-INFORMATION INTERACTION
Accomplishments and Advancements
HII is a new program in the Information Sciences Campaign, and had been in operation for about one year, bringing together researchers from disparate disciplines and technical backgrounds. “Interaction” is what distinguishes HII in the information science and human science research space. It applies
to the interaction of groups of individuals and agents with information, as well as individual discovery, creation, and use of information. Research into interaction requires expertise in collaboration, team dynamics, social influence, and other social science subareas. Interaction is at the center of battlefield dynamics, the expanding use of social networks for communication, and information dissemination, and it enables the dismounted soldier to attain and maintain extensive situation awareness. With a focus on the dismounted soldier, issues of land-based real-time interaction rather than future forecasting are paramount. The objective of HII research at ARL is to develop models, methods, and understanding of data and information generated by humans and intelligent agents in a complex, multigenre network environment. It further examines tools to respond to user information needs with due consideration of user variability and mission constraints, and thereby to develop timely and accurate situational understanding.
Multimodal State Classification
The research addresses an important problem of human state classification. It focuses on the use of machine learning to classify whether a user is in a threatened or challenged state based on wearable sensor data. The research focus resides in comparing and contrasting multiple modalities and multiple methods of classification, and provides breadth to the research while ensuring broader strategic relevance. The potential of the research is high, and the investigator demonstrated understanding of how this work fits into the broader scientific community.
Exploiting and Mitigating Cognitive Effects of Real-World Luminance Dynamics on Visual Search
This research explores the use of deep neural networks to predict human search behavior under dynamic range stimuli for images and video by illuminating the complex patterns of interaction that impact search. Specifically, the focus is on developing models of visual search behavior in dynamically changing, high-luminance environments such as those available in high-definition resolution imagery and videos. The work is in collaboration with the Human Research and Engineering Directorate at ARL.
Assessing Value of Information for Graph Understanding
The focus of this work is in developing metrics to measure the fitness of statistical graphs that better inform the design of Army displays, and to develop automated techniques for generating task-dependent statistical graphs. There exist some human-factors guidelines for generating graphs and charts, and the use of these guidelines has been shown to yield improved performance. The work has clear alignment with Army needs and represents collaborative work between ARL, the Naval Research Laboratory (NRL), and the Institute for Human and Machine Cognition.
Social Computing: New Directions for Army Relevance
This is a new area of research and is focused on applications in decision making, in establishing context, and in understanding or interaction. The scope of research tools and technologies is quite broad, and it will require input from many disciplines. The program is in a formulation stage, seeking to establish appropriate research directions in social computing relevant to Army needs, and to coordinate disparate efforts in the field across ARL. Machine learning and graph theoretic analysis have been identified as key technologies to pursue in this context.
Intelligent Information Management for the Battlefield
This user-focused research is of high quality, has a clear articulation of the associated scientific and technical challenges, has clear connections to Army needs, and presents a transition path and partner. The researchers demonstrated strong understanding of what work is being done in the broader scientific field and how their work fits into that framework. As such, the researchers know what community to address and where to publish.
The work is creative, and the data and methods are appropriate to the objective. The researchers were able to gather and make clear utilization of end-user feedback. A strength of this work is the use of live experiments to gather data, shape the research objectives, and test the overall approach. The demonstrations and live experiments provide a test-bed and framework for VoI research that lays the groundwork for other related projects in the HII area. This work has the potential to develop context, user-adaptive information management, and processing techniques that will support the dismounted soldier in tactical environments.
Opinion Formation and Shifting
This project seeks to understand how people passively interact with social media, and will contribute to a theory of information propagation through media-enhanced social networks. The researcher is very talented and has good ideas. The collaboration within ARL is good. Using a set of experiments, the investigator is seeking to derive empirical thresholds for opinion formation and shifting given diverse types of media. This is clearly of Army relevance. The investigator has solid scientific credentials, is good at designing experiments, and would be well served by mentors from the area of sociology or social psychology of opinion formation.
Internet of Battlefield Things
The complex cyber physical environment that will characterize the battlefield of the future has defined the Internet of Battlefield Things (IoBT). The vast amount of information available from networked things in IoBT offers endless possibilities of using analytics to support military planning and operations. This will require humans to interact with intelligent agents in a variety of ways. The research project seeks to understand the dynamics of human-agent interaction, and to build models of such interaction so as to simulate different battlefield scenarios to develop optimized strategies for dynamic resource and task allocation across humans and agents in IoBT. Toward this end, the approach is to formulate an optimization problem based on a cross-layer (Com/Net, actuator, human layers) mathematical model for optimal resource allocation. This project has clear Army applications and seeks to collaborate with researchers in the NRL. Developing constraints reflective of human knowledge and perception will certainly enrich the optimization problems, but the approach for doing so was not presented.
Challenges and Opportunities
The HII agenda at ARL is very broad, and without a dramatically narrowed focus, it will be difficult to make sustained progress that will transition to Army-relevant applications. A key challenge, therefore, is to leverage the deep understanding of Army needs and requirements to narrow the focus and to reorient new projects. In a number of the projects, the research questions are too broad, and are not designed to lead to actionable and transitional results in the near term. In cases where research is
linked to Army needs, the specific Army issue is not clearly articulated in a way that drives the research question and the methodology. The approaches used are often not well matched to the available data or to the research question. A strong scientific mentoring program is needed to ensure that promising projects realize their potential.
In response to suggestions from the 2015-2016 ARLTAB assessment, ARL has recruited two psychologists into the HII program and is pursuing the recruitment of an economist. Although this initiative is to be lauded, it is important to recognize that talent with necessary scientific expertise in the areas of social, team, and group behavior and communication is needed to support the HII projects. Data scientists and information scientists have been hired, further strengthening the computer science focus. The research staff distribution continues to require additional balancing to bring greater understanding of social theoretical underpinnings of the research.
HII has begun to identify synergies with other parts of the ARL, and may benefit from creating a “social network model” of ARL, identifying those individuals whose work is synergistic with their own. There is untapped potential to coordinate across areas concerned with the social side of human behavior, and to interact with other agents and information. The creation of a challenge problem could be a way for focusing HII efforts. Although a challenge problem has not been identified, HII researchers are engaged and active in discussing what one might be. Mentoring, particularly matching researchers to others with scientific experience in the methods or theories of relevance, needs to be a priority for ARL.
From the work that was reviewed, it appears that there are three areas of research around which multiple projects can be organized, and where closer ties among projects would be a force multiplier. These areas are VoI, social media and diffusion, and agent-based modeling. There is a strong focus on VoI, and this is the most mature of the concentration areas. Projects related to this area could develop and adopt a common use case, share data, and develop common formulations of VoI. The HII research needs to expand to recognize the impact of context. Whether the information is of a sociocultural origin or technology related, the VoI is contextual. Taking context into account, such as the physical climate and social environment, is a major challenge and opportunity. A focus in this arena has the potential for elevating HII as leaders in the Social Computing, Behavioral-Cultural Modeling, and Prediction and Behavior Representation in Modeling and Simulation (SBP-BRiMS) community.
The other two areas are broadly part of social computing. Social computing brings in research from over 20 scientific disciplines, most of which are not represented in HII at ARL. Despite the commendable step of organizing a workshop, the HII group is still at a beginning point on the learning curve in this area. Social computing is an area that is growing exponentially. The HII group is beginning to embrace a small sector of this community, and it could task itself with monitoring the larger social computing field as it develops. To stay actively engaged, the group could identify Army-specific problems that it needs to pursue and invest in the personnel qualified to pursue the work.
Social media and diffusion of information is of interest to HII, specifically as it relates to the role of social media in information diffusion and propagation of false information. This field is extensively researched outside ARL, with a decade of research addressing issues that need to be incorporated into current HII formulations. Social media presents both an opportunity and a challenge, and for HII to be a meaningful contributor, there is a need to define a uniquely Army mission, identify a transition partner, and understand the legal authorities that constrain Army activity in this space.
The projects making use of agent-based modeling have strong potential but need to be more tightly focused on Army needs. The value of agent-based models (ABMs) is that they support prediction and forecasting. Since HII is focused more on near-term analytics rather than on forecasting, the aspect of ABM that supports forecasting is not needed. Another value of ABMs is that they support theory
development, and can be used to fine-tune field experiments. This aspect of ABMs has strong potential for the HII program. The ABM expertise within HII needs to be enhanced to achieve greater progress.
It is important to underscore some disconnects that may detract from the impact of HII research. First, there is a disjuncture between work being done, the stated goals and objectives, and the perceived time-to-get-to-that-objective. As an example, HII goals include operation in a complex multigenre context and understanding information content. Much of the research, however, considered only simple or single-genre contexts and ignored content. Another example relates to the time frame allocated for a research task. Social dynamics is meant to be a “distant goal”—that is, one that will take more than 15 years. However, some HII researchers working in that area anticipate success in 1-3 years. This would be acceptable if the ongoing work was research needed to enable the longer term goal; that does not, however, appear to be the case. Shifting to the research program organization, it is to be noted that the HII program overview mentioned “beliefs” multiple times, but the specific work that is most closely associated is focused on trust and VoI. Trust and VoI are not the same as beliefs, particularly as they have been operationalized. Either the goals need to include trust or the research could be shifted to beliefs.
To accelerate progress in the crucial areas of team dynamics, collaboration, and social influence, ARL could invest in additional social scientists to balance its strengths in engineering and computer science. Specifically, a mathematical sociologist and an experimental social psychologist could help significantly with ongoing and proposed work. A systematic effort to survey this expanding area to identify specific Army-relevant subtopics and research results, and the groups working in this area, could also be a priority.
The HII group could develop one or two challenge problems. To support the challenge problem(s), the HII could consider a unifying integrated test and evaluation platform that is relevant to that challenge problem. This would support building toward strongly Army-relevant technology, and support the testing and evaluation of basic work. For every project, the researchers could articulate how this supports the deployed dismounted soldier, and be able to pinpoint the specific Army relevance, the Army issue, and the research they can do that is not being done elsewhere.
As noted earlier, the scope of the research program in HII could be further narrowed and refined for maximal impact. One possible area for focus is a framework that defines how a joint human-machine system selects a level of data resolution for effective interaction. A key issue in this task is to understand what role the human operator needs to play in this selection and what is best relegated to automation. There is no theory or framework to address this problem in a complex dynamic environment with teams of humans and autonomous systems. A good theoretical understanding of this problem will be important, because in the long run, the Army will need to transfer data at different levels of resolution for cultural information, social network information, and general social informatics. HII has the opportunity to establish leadership in core scholarship in this arena.
A second area to consider is a concerted focus on problems involving teams of three or more interacting agents (e.g., two humans and one artificial). Currently, several HII researchers are focusing on problems related to one human and one artificial agent. Fundamental questions to be answered in larger team interactions include what enables high performance, how much information overlap is needed, and what kind of transactive memory and knowledge memory they need to develop. These questions assume increased importance as the team size is increased, or as team members come from different services or from different countries. Additional scenarios need to be considered where information cannot be shared with all team members, and where there is impact of these omissions on performance. Another fundamental question to be addressed is how the mode of communication—video, image, voice, and so on—impact team performance.
A third area for consideration deals with the issue of the commander’s intent, and how intent can be transmitted so that it does not get distorted. It would be important to understand what factors cause distortion, and how coordination and interaction among soldiers help resolve ambiguities in their understanding of the intent. Much research in HII could be refocused to address issues directly related to commander’s intent. A platform could be built for testing and evaluating models and conducting experiments vis-à-vis commander’s intent in a multigenre complex information environment.
Multimodal State Classification
This research could be strengthened, particularly with respect to methodology. There was lack of a clear articulation of connection to an Army mission, and the associated lack of a clear transition path and transition partner. This project could benefit from the combination of mentoring in the broader scientific community, and advice from potential end users and colleagues to help identify the specific Army focus and transition path. While the initial results indicate success in working with physiological data, the likelihood of it working with more cognitive or social factors is in question. The project could benefit from using much larger training sets.
Exploiting and Mitigating Cognitive Effects of Real-World Luminance Dynamics on Visual Search
Initial results demonstrate feature interaction, but no link of this to in-the-field utility was demonstrated. It was unclear if there was sufficient data for model development and testing, especially in the context of using deep neural networks. Another goal of the project is to develop a high-definition resolution model of visual saliency, and it would be useful for the researchers to review the literature on how saliency has been addressed in previous work.
Understanding Theoretical Information Interaction: The Development of a Standard Model Using an Agent-Based Modeling Framework
The objective of this research is to address challenges in developing fundamental theories of human information interaction. This project is central to HII, but it needs to be better formulated. The project is relatively new, and both the domain and the methodology are new to the investigators. While the investigators are creative and energetic, the goals of the project are overly ambitious, and the team has limited expertise in agent-based modeling. A project of this scope and the development of a credible agent-based model in this space could take years and would employ vast quantities of data for model instantiation and testing.
There are many extant theories that are considered to be theories of human information interactions. These include social influence theory, constructuralism, information bias, and so on. The presenter had knowledge of a few of the behavioral theories such as confirmation bias. The approach being followed is to develop an agent-based model where complex interactions can be evaluated. However, there are multiple existing agent-based models that have been used to reason about human information interaction. It is important for researchers to learn about these models, and that the model being developed be docked against these existing models. A simple model in NetLogo is being developed that is great for learning but will not scale well for large simulations. This model does not control for structural effects such as the networks connection information, the social networks connecting people, the communication networks over which people interact, and the knowledge networks of who knows what, or has access
to what information. The authors may consider employing a meta-network approach to look at high-dimensional networks. The work could also benefit from the use of agent-based modeling tools that allow for network analytics. Furthermore, the project could be strengthened by building on the measure of VoI developed elsewhere in HII and using that as one of the dependent variables in the simulation. That would both increase Army relevance and create synergies with other work in this area.
Assessing Value of Information for Graph Understanding
This project has a well laid out plan but is limited in that it does not have a comprehensive theory, or empirical mapping of the relation between graph type, decision problem, and data. Without this mapping, the overall guidelines developed are likely to be too generic.
Social Computing: New Directions for Army Relevance
This scope is somewhat narrow, and reflective of a strong computer science focus. Social computing is a well-established area, involving many scientists and multiple conferences devoted to the subject. The investigators present a description of social computing as dealing with social media but demonstrated limited understanding of the breadth of work in that field and a grasp of the key challenges.
Value of Information-Based Mission Needs Matching
The objective of this research project is to evaluate a set of metrics that could be used to assess value of information within a simulation environment. This is a relatively small project that uses existing metrics with clear Army relevance. The project could be expanded to other metrics and the simulator expanded to other types of conditions. This, and indeed all the VoI projects, would be strengthened by integrating all the simulation and test-bed environments, creating a comprehensive set of VoI metrics, and utilizing consistent approaches to measuring and calculating them. Tighter integration across all the VoI projects would be a force multiplier. The simulator and the metrics are really best for evaluating physical systems, and as currently developed, will not work for sociocultural information. Indeed, such sociocultural information is difficult for all the VoI projects. Furthermore, the current simulator is limited in the kinds of missions that it can accurately reflect. The authors might consider developing a taxonomy of mission types so as to accurately gauge where the current simulator is relevant, and where new simulators are required. It is proposed that the work be extended to metric aggregation. This is a worthy goal, and the researchers could build better awareness of aggregation techniques cited in the literature.
Modeling and Analysis of Uncertainty-Based False Information Propagation in Social Networks
This work proposes an opinion model based on subjective logic and then uses it to study how to mitigate the impact of false information using counter-narratives. The model is mathematically well specified but operates at a very abstract level. The problem is important, and the researchers need to get a better understanding of other ongoing research related to fake news and rumor propagation. The work is directed primarily at spread of false information through social networks, but the model does not take into account known network effects. Further, the emphasis is on thinking of networks as graphs. The vast literature on information movement, fake news opinion dynamics, and social change in networks that
has come out of the social sciences is very relevant, and needs be considered. The researchers also need to dock their model against leading models in this area that are based on social influence. They would also benefit by gaining familiarity with the broad literature on information diffusion, misinformation, and fake news. Key factors not considered are the psychology of opinion formation, social influence, constructuralism, network topology and its impact on diffusion, network externalities, information biases, and confirmation biases. This is a highly interdisciplinary problem and requires expertise beyond the discipline of engineering. The researcher is quite talented and would benefit from going to the more social science conferences in this area—such as the International Network for Social Network Analysis (INSNA) meeting or SBP-BRiMS.
Opinion Formation and Shifting
The principal weakness of this project is a lack of consideration of a broad range of existing work on opinion formation and shifting. Specifically, this work is classifying people based on number of messages by type of media, with no attention to content, saliency of content, relation of content provider to the person being influenced, and so on. Prior work has demonstrated that it is not solely the number of messages that matter, but rather factors such as the affective relation between the sender and receiver, the content of the message, the affective nature of the message, the inclusion of image/video/numbers, the role of the sender in the receiver’s network, and the strength and direction of the receiver’s current opinion. Another issue arises from the claim of obtaining results using “content-laden message examples,” but content is explicitly removed in the experiment. This is clearly inconsistent. Use of AMT for subjects is also problematic. It is not clear that the subjects were sufficiently tested for linguistic skills commensurate with the project and for the learner type. Individuals have different ways of learning, and some are more prone to learning from video and images than others. This needs to be controlled in the experiment.
ATMOSPHERIC SCIENCES
The research portfolio of the Battlefield Environments Division (BED) can essentially be divided into two thrust areas that include improving environmental understanding of the planetary boundary layer (PBL) and processes that operate on small spatial and temporal scales, and on developing appropriate environmental intelligence tools for deployed soldiers to use in austere, complex operating environments. The review in this cycle followed on the heels of an assessment held at White Sands Missile Range (WSMR) in June 2016. Thus, this interim report spans a period that is one year instead of the two-year cycle of reviews that was adopted in 2013-2014.
Accomplishments and Advancements
BED scientists have accomplished a great deal, and these achievements will be highlighted throughout this section. Additionally, BED leadership provided specific, point-by-point responses to the findings and recommendations from the 2015-2016 Army Research Laboratory Technical Assessment Board (ARLTAB) report, and greatly assisted the panel in its work.
The research projects reviewed by the panel during this cycle included detection and characterization of chemical aerosols, acoustic and infrasound sensing, development and fielding of a meteorological sensor array at WSMR, and advances in small-scale atmospheric model development, verification, and validation. In addition, updates were provided to the panel on past large-scale projects. These are described shortly.
It is important to note that additional projects were also discussed with the panel. Significant progress has been made in the past year with the automated aerosol Raman spectrometer project. The work is commendable, and a recent paper1 is particularly useful in showing the utility of the project as well as discussing the benefits of the BED work over other aerosol characterization systems. Additionally, work that was completed through the Small Business Innovation Research (SBIR) program to improve drone performance is also noteworthy. This innovative leveraging of the SBIR program has been successful in developing enhanced drones that incorporate environmental turbulence data, thereby improving drone movement. The SBIR work that has been undertaken by the BED has multiple civilian and DoD applications and, most importantly, has led to the development of additional, fundamental (6.1) scientific research questions.
Design of Experiments for Verification and Assessment of Fine-Scale Atmospheric Forecasts
This presentation was essentially an update of the team’s activity during 2016-2017. Good progress has been made regarding the development of an appropriate design of experiments (DoE) matrix for initial testing. The matrix, developed for 40 weather research and forecasting (WRF) model simulations, was recently completed and is ready for execution. The significant challenge of developing a DOE approach for atmospheric numerical weather prediction (NWP) models, an approach that has never been attempted, had been noted by ARLTAB earlier. The ARL team has done an admirable job of surveying the scientific community to evaluate the project’s feasibility and establish a unique niche for this work. Much work remains to be done, particularly with model verification. At present, the team proposes to employ root-mean-square error analysis of the 40 model runs, with the intention of performing more detailed verification using subdomains, and possibly leveraging the National Center for Atmospheric Research (NCAR) method for object-based diagnostic evaluation (MODE), which is currently being evaluated at ARL.
Multiscale Atmospheric Modeling for Tactical Army Nowcasting
The researchers have continued the performance evaluation of the WRF model for the May 2013 Moore, Oklahoma, tornado case. The model was tested in 2016 with a number of different data assimilation schemes, including a hybrid of four-dimensional (4D) data assimilation and three-dimensional (3D) variational data assimilation that is both computationally efficient and fairly accurate. Researchers have been running a very-large-eddy-simulation (VLES) version of WRF over the Jornada Experimental Range (JER) region north of WSMR, with a 300 m horizontal grid spacing and 130 vertical levels. Preliminary results from the VLES WRF show great promise in the realistic reproduction of detailed PBL vertical motion patterns over the JER region. Results from a version of WRF run on a 450 m horizontal grid over the Dugway Proving Grounds were also presented. This set of simulations sought to evaluate combinations of PBL and surface physics parameterizations in order to determine which combinations produce the most accurate representations of the PBL in complex terrain.
___________________
1 D.C. Doughty and S.C. Hill, 2017, Automated aerosol Raman spectrometer for semi-continuous sampling of atmospheric aerosol, Journal of Quantitative Spectroscopy & Radiative Transfer 188:103-117.
Vortex Filament Method in Microscale Atmospheric Modeling
Researchers have pursued three distinct approaches for the atmospheric boundary layer environment (ABLE) model for predicting mean wind, temperature, moisture, and turbulence over urban and complex terrain in near real time. These included computational fluid dynamics, the vortex filament method (VFM), and the lattice Boltzmann method (LBM). There are two attractive characteristics for the VFM that include its potential use as an engineering tool to help one understand the nature of complex terrain and urban-scale atmospheric flows, and the validity of the VFM approach in high Reynolds number scenarios. Preliminary results feature the ability of VFM to reproduce thermal bubble rise by comparing simulations to theoretical results from Shapiro and Kanak (2002),2 and demonstration of isotropic turbulence in a 3D periodic box.
Atmospheric Boundary Layer Environment Model Lattice Boltzmann Method
The LBM approach solves particle kinetic equations in lattice directions, and is well suited for modeling turbulent flow with complex boundaries. It is a more applied method than the VFM discussed previously, and has thus far satisfactorily reproduced theoretical results for stably stratified flow over a mountain ridge, convective flow forced by a heated ground surface, and flows around buildings. It is worthwhile to note that this project and the VFM have benefited greatly by collaboration with researchers from multiple universities and other ARL directorates.
Visualizing Terrain in Augmented Reality
This project relates to 3D visualization of terrain and ground cover and is illustrative of collaboration between BED and other Information Sciences Campaign researchers. The terrain and ground cover data were collected from UAVs and could be visualized using either 3D glasses with a projection screen or a holo-lens wearable visor. Both methods were demonstrated in the poster presentation. ARL collaboration on this project with the University of Southern California Institute for Creative Technologies is noteworthy. The project has clear Army relevance, providing soldiers with the ability to plan and rehearse dangerous missions in complex terrain and urban areas. This technology also has potential applications in a training environment.
Infrasound Research and Applications
This more-than-15-year applied research (6.2) project is enhancing techniques for detecting blast signals using infrasound waves—typically at frequencies below 20 Hz. These waves can travel for hundreds of km horizontally and vertically, making them ideal for standoff detection of events such as missile launches. The approach uses time-of-arrival (ToA) differences of sound, and incorporates these differences with atmospheric properties to improve the models to estimate ToA. This work is critical to the proper identification of activities and their locations. The research incorporates real-world measurements of infrasound ToA from rocket launches from Wallops Island, Virginia, under varying conditions; uses models with the appropriate parameters to simulate the actual systems; and then compares the actual measurements with the modeling results. This project therefore incorporates an excellent work plan to
___________________
2 A. Shapiro and K.M. Kanak, 2002, Vortex formation in ellipsoidal thermal bubbles, Journal of Atmospheric Sciences 59(14):2253-2269.
validate the utility of the models. One of the unique aspects of this project is the relationship between ARL and the recipient of the technology, the Army’s 501st Military Intelligence Brigade in the Republic of Korea. While the technique in its current state is not ready for field use by the brigade’s soldiers, ARL is currently involved with the brigade in testing and evaluating the methods. A clear transition plan to different parts of the Army is in place, and includes, among other aspects, collaborations with international partners (e.g., Republic of Korea). The work is strong, and improvements to the mathematical modeling (to improve the time-delay estimates) are possible through collaboration and connection with researchers in the sensing and effecting group.
Acoustic Classification Algorithms for Complex Propagation Channels
This project uses the modeling of sparse data sets to model acoustic events with the intention of developing event classifiers that can be used by soldiers in the field. Work was conducted at the basic research (6.1) level and is in the final year of 6.1 funding; it will transition to applied research (6.2). The work technically represents a joint battlefield environment effects/sensing and effecting project and its science is solid.
Detection and Characterization of Chemical Aerosol Using Single-Particle Laser Trap Raman Spectrometer
This project has successfully demonstrated a laser-based technique for isolating, detecting, and identifying the chemical compositions of micron-size particles of multiple phases with unprecedented speed and accuracy. The researchers’ method to capture and suspend the particle vastly reduces the analysis time to the order of seconds, and allows a sampling frequency that is much superior to other methods that are currently in use. The work is exceptional and novel, and it has the ability to revolutionize the aerosol science field as well as all industries and technologies that rely on aerosol science. The unique feature of being able not only to measure composition but also to track changes in the size of the particle paves the way for new investigations and applications, and is therefore transformational to the aerosol field. As one example, the work may assist in developing basic information on the aging of particles, thereby helping to determine their origin. This project has the potential to spawn additional applications outside DoD, and once it has been adapted to an operational environment, provide soldiers in the field with an additional level of protection from chemical weapons. Several papers on this work have been published, are in the review process, or are being planned for submission to high-quality peer-reviewed journals.
Meteorological Sensor Array
The deployment of the meteorological sensor array (MSA) at WSMR will enable unprecedented continuous examination of atmospheric phenomena crucial to our understanding of atmospheric flows over complex terrain at high horizontal resolution. In addition to fixed instrumentation, it also a mobile component (lidar) and instrumented UAS. As a means of laying the groundwork for this ambitious project, the ARL team has been very engaged with universities, industry, and international agencies in field observation experiments such as mountain terrain atmospheric modeling and observations (MATERHORN), and an experiment that is just concluding in Perdigão, Portugal. The deployment of the MSA has four phases. Phase I is the experimental test-bed of 5 towers located near the WSMR Las Cruces gate. These towers have been operating as needed for the past 2 years. Phase II involves the installation of 36 meteorological sensor towers at JER adjacent to WSMR, and a memorandum
of understanding between ARL and the U.S. Department of Agriculture (USDA), which operates the National Wind Erosion Research Network in the JER, has been established. Phase II instrumentation deployment in USDA’s JER data collection network is more than 50 percent complete. There are also plans for an acoustic array and instrumented UAS over this area. Phase III will involve a dense instrumentation network installation on the west slope leading up and over San Andres Peak at 2510 m elevation to complement the ARL/USDA network over JER. This capability, when fielded, will provide detailed atmospheric information over this extremely mountainous terrain, with resulting data certain to benefit the Army’s ability to operate in such challenging environments. In Phase IV, ARL hopes to gain approval for installation and operation of instruments in the High Energy Laser Systems Test Facility domain, which is part of WSMR. To date, nearly a dozen agencies and universities either have provided equipment or are interested in collaborating with ARL on this project. For instance, NRL is contributing soil moisture sensors and rain gauges to the MSA for verifying satellite-based moisture measurements in a desert environment.
Challenges and Opportunities
Current staffing in BED is 53, and includes 45 government personnel and additional contract personnel including two Ph.D. senior researchers, two postdoctoral fellows, two computational support staff, and two administrative support personnel. These numbers reflect a sharp decline in BED contract personnel over the previous year, in both research and support staff. The engagement of postdoctoral fellows has greatly benefited the quality of research within BED, and the decline in their number is viewed with concern. The engagement of postdoctoral fellows has been shown to be effective in the past, and needs to be continued, especially if promising candidates can be converted to permanent staff. A new ARL initiative is the establishment of nine ERAs that emphasize interdisciplinary efforts considered essential to the success of Army operations in the 2030-2050 time frame. BED researchers have been quick to show the linkages between their individual projects and the appropriate ERA. BED can be integrated within additional multiple ERAs.
The challenges and opportunities to BED programs are interrelated. First, BED’s thrust areas touch nearly every S&T campaign outlined in the Army Research Laboratory S&T Campaigns 2015-2035. Although battlefield environment and weather are technically assigned to the sensing and effecting area within the Information Sciences Campaign, the atmosphere and its effects can impact vehicle maneuver, lethality and protection, human sciences, and materials science—all of which are campaign areas in the ARL S&T plan. A related issue is the continuing challenge that BED faces to ensure that its unique expertise is being leveraged by ARL research projects that may have an environmental sensitivity. Both of these represent opportunities for additional research collaboration that can enhance the overall strength of the ARL portfolio, and strengthen its position for transitioning useful technology to the field. On the other hand, these interactions are also challenging because many project scientists outside BED are not aware of the expertise residing in the directorate, and they may not even understand that atmospheric effects and impacts need to be incorporated into their research. This latter point was illustrated in several projects from the system intelligence and intelligent systems and sensing and effecting areas, where the researchers either were not aware of the need to incorporate environmental sensitivities into their research methods or felt that environment was unimportant or could be mitigated by other strategies.
BED has continued to leverage all available resources in advancing the research plans. In addition to postdoctoral fellows, the division has made extensive use of other avenues for assistance on its projects. For example, students from Navajo Technical University are at ARL as summer visitors. Another instrument that BED has utilized is the cooperative research and development agreement (CRADA).
Planned future CRADAs with NCAR and Vaisala Corporation will help all three BED branches work more efficiently—with the Vaisala CRADA being particularly important as the MSA is fielded. However, there are some inefficiencies regarding the length of time to get CRADAs established. In an environment with limited resources, this presents a challenge to maintaining research at the leading edge. BED continues to be challenged by steadily declining resources, but there is a strong scientific imperative to incorporate BED expertise in multiple projects across the ARL portfolio. If full collaboration potential with other related laboratory projects can be realized, fundamental science will be advanced in highly significant and potentially groundbreaking ways.
Multiscale Atmospheric Modeling for Tactical Army Nowcasting
Challenges here include a more complete understanding of objective verification statistics for radar reflectivity, where the current equitable threat scores are somewhat counterintuitive. The preparation for eventual use of data from the MSA being fielded at WSMR was also discussed. The researcher commented about the potential of leveraging findings from the Design of Experiments project in future research once that project has produced usable results.
Atmospheric Boundary Layer Environment Model Lattice Boltzmann Method
Future work includes improvements in turbulence model coupling as well as radiation model and surface thermal model coupling, continued validation using laboratory and field observations, and initialization with larger-scale model data and lidar observations.
Visualizing Terrain in Augmented Reality
The researcher was directed to a study published by Hembree et al. (1997)3 that incorporated cloud scene simulation model output into a mission planning and rehearsal simulation tool called PowerScene. This early work was limited by hardware computational capabilities, with the simulation tool losing its real-time frame rate as a result of assimilating the weather model information. Significant advances in computational power render the approach more tractable, with potential for realistic weather incorporation into contemporary simulation tools.
Acoustic Classification Algorithms for Complex Propagation Channels
The departure of a key BED researcher on the project has resulted in the modeling effort not sufficiently incorporating atmospheric effects in the physics of acoustic propagation. The current approach is to mitigate or overcome the effects of the atmosphere through deployment of sensors in multiple locations. However, the project could benefit from increased collaboration with other BED subject-matter experts for an improved integration with the atmospheric components.
___________________
3 L. Hembree, S. Brand, W.C. Mayse, M. Cianciolo, and B. Soderberg, 1997, Incorporation of a cloud simulation into a flight mission rehearsal system: prototype demonstration, Bulletin of the American Meteorological Society 78:815-822.
NETWORKS AND COMMUNICATIONS
Accomplishments and Advancements
The research portfolio of the networks and communications area concentrates on understanding and exploiting interactions of information with sociotechnical networks, particularly communications, and command and control networks. The research comprises three broad topical areas: channels and protocols, control and behavior, and information delivery. Human-machine teaming is a growing topic in all three topical areas.
Channels and protocols consider challenges in the transfer of information between nodes in an efficient and robust manner. Army-specific needs focus on protocols that address network diversity, robustness, and survivability. ARL researchers are exploring a variety of new heterogeneous approaches that take advantage of heterogeneous channels and supporting protocols to address Army requirements to operate in challenging terrain and atmospheric conditions, and in the presence of sophisticated adversarial disruptions.
Research in the area of control and behavior is focused on the complex dynamic network and environment behaviors and needed network controls to support Army-relevant networked communications. These environments include social networks, physical networks, and networks that include coalition partnerships. There is a need for adaptive protocols and self-adaptive networks that can be resilient and provide effective networking operation in a complex and rapidly changing environment.
Information delivery includes the study of methods, theories, models, and properties for storage, management, transformations, transmission, and delivery of information that is mission-relevant and context-appropriate through socio-technical networks. Essentially, the goal is to meet the information needs of the soldier in mobile, tactical environments. The research includes network-based information processing and nodes and architectures that support combined communications and sensor processing. Of interest is the ability to construct reliable and secure information networks that incorporate an understanding of human behavior and constraints due to the terrain. Low power and low bandwidth are critical. Distributed, user-oriented, multiscale network summarization and analytical processing solutions are needed, as are factors and methods for enhancing human’s and agent’s trust in received information.
Since the last review in 2016, ARL scientists and engineers have made significant progress in many of these areas of research.
Event Tracking from Streaming Open Source and Infrastructure Data
This basic research project was in its fourth year of a five-year program. The research focuses on modeling of data streams and fusion methodologies, pattern of life assessment, and applying semantic reasoning to detect potential new outcomes of activity or to look for anomalous activity. Recent accomplishments include the fusion of sensor and text-based data. The research considered two sources of data: social media data collections from individuals, and physical tracking of entities (people, vehicles) in a geographic region. The use of social media data is a good first step, and more sophisticated data sets and social media models are available in the research literature; future work could consider the role of adversarial actors as potential corruptors of social media data. The research focus demonstrates an understanding of Army-specific requirements and technical gaps. The researchers understand the need for additional work, including how to scale algorithms and processing. They also demonstrated good understanding of the sensor data and its limitations, and the need for appropriate data-generation models.
Robust Distributed Optimization with Dynamic Event-Triggered Communication
This basic research effort is aimed at developing distributed implementation and communication strategies that would apply to tactical inference tasks that include distributed inference or learning. The research shows that under mild conditions, the algorithm can perform robust fusion at lower communication cost than a selected optimal algorithm, achieving optimal inference performance at the cost of higher delay. The research shows a good understanding of current work and Army needs.
Tactical Optical Communications and Hybrid Communication Networks
The overarching motivation for this research is the need for heterogeneous communication schemes that address the challenging environment of tactical communications. Recent work has focused on heterogeneous communications that couple traditional RF communications with deep-ultraviolet (UV) light communications. The UV communications both provides non-line-of-sight communications and facilitates covert and jam-resistant operations. The research represents a good blend of theoretical modeling to understand fundamental UV communications properties and an experimental test-bed demonstration to show the proof of concept of both UV communications and heterogeneous RF-UV communications. Spatial multiplexing and code division multiple access to support multiuser communications were also considered. Overall, this is a very strong and well thought out research program that is developing fundamental understanding of UV communications, and validating this understanding via field measurements. Clear transition plans are stated, and the researchers show a very good understanding of Army-specific needs and system requirements. The program has strong leadership, and early-career scientists in the program appear to be receiving good mentorship.
Low-Power, Low-Frequency Mobile Networking
This program is focused on developing low-frequency RF communications and networking capabilities that have practical applications in Army tactical deployments, especially in urban terrain. The research includes both theoretical modeling and prototype designs for testing and validation. The research includes highly innovative design of very small antennas (1/50 wavelength in size) that enables the deployment of novel radios on small mobile ground or air vehicles. As compared to a standard passive antenna, the proposed design has tripled the bandwidth, enabling practical communication rates. The communications techniques, including code design and performance bounds derivations, are impressive. The system designs and their theoretical analyses are impressive, and the project demonstrates system-level understanding of Army needs. This is a strong and comprehensive research program that is expected to result in Army-relevant solutions. The researchers need to develop clear paths for transitioning the research into deployed systems—for example, through the U.S. Army Communications-Electronics Research, Development, and Engineering Center. The researchers could also consider patenting their active antenna if they have not already done so.
Networking in Resource Constrained Environments
This is a large project that began in fiscal year 2010 to develop networked communication devices and network protocols that meet Army needs of energy-efficient communication systems that operate under a wide variety of conditions. The program has produced a steady improvement in system throughput and energy efficiency that has significantly improved performance and network protocol understand-
ing. Recent advancements include developing a network protocol that minimizes the battery drain on tactical radios, developing small-size and low-weight and -power radios that integrate radio frequency and UV communications channels, and developing heterogeneous networking protocols. The prototype system testing in a military environment (Camp Roberts, California) to demonstrate self-healing networks and to verify the low energy consumption is a laudable accomplishment.
Quantum Networks and Robust Quantum Channels for Photonic Entanglement Distribution
This project represents an important component of ARL research in quantum networks that began in 2015. The program is part of a larger collaboration within DoD to develop both a theoretical and an experimental capability to understand quantum communications systems. The researchers were able to clearly identify healthy collaborations with both academic researchers and other DoD programs. System-level understanding and long-range communications have been identified as niche areas for ARL; this complements and leverages component-level work at universities and materials and device development programs elsewhere within ARL and other institutions. While early in development, the program currently has six projects that seek to understand and address system-level needs necessary for developing viable quantum channels and networks of military relevance. There is a good mix of theoretical and experimental projects, and the ARL quantum networking test-bed provides a valuable tool for validating theoretical predictions. Overall, the program is making valuable contributions at this early stage.
Geometric and Topological Structures in Composite Networks
This research project involves a mathematical treatment of dynamic or evolving networks, with a focus on geometric and topological methods. The focus is on adapting novel mathematical concepts like curvature and homology to network science, through an empirical analysis of real network data. The research appears to be of high quality, and the researcher is both well-qualified and collaborating with external researchers in industry and academia. The application of strong collapsing theory to hole localization and sparse coverage in location-unaware sensor networks, as well as the use of topological data analysis for network classification, appears promising.
Practical Security for Tactical Networks
This program focuses on information-theoretic techniques for provably secure communications. The goal of this effort is to provide authentication and secrecy in resource-constrained environments of tactical networks by relaxing the requirement of complete information security, thereby reducing communication resources such as bandwidth or computational needs for information decoding. An achievable region of data rates versus secrecy and authentication has been derived; this region is higher than previously known and the methods are included in an invention disclosure. In addition, a prototype implementation on software-defined radios was described. The researcher seems well-qualified to carry out the research program, and the results are of high quality and are being published in appropriate venues. The models and proposed approaches include consideration of operating environments that are relevant to Army missions. A clearer indication of practical implementation of the research in Army operations would be helpful.
Software-Defined Networking for Tactical Networks
This small basic research program is a new start since the previous review. The program seeks to develop a robust experimental environment for testing of decentralized software-defined networking (SDN) concepts. The program specifically seeks to address the vulnerability in military networks of a centralized SDN control of network behavior, and proposes a framework of distributed SDN control. The experimental platform provides a helpful test-bed to test network control using measured and synthetic data that mirrors Army-relevant tactical networking challenges and constraints.
Challenges and Opportunities
Much of the networks and communication work reviewed showed both a clear problem statement and a clear justification of Army relevance, but this was not always the case. There is an opportunity to enhance the impact of the work if researchers clearly articulate the Army needs and how the research fits in with larger projects or campaigns. The taxonomy of identifying, for larger projects, what research elements ARL intends to lead, collaborate, or follow, was found to be very helpful in assessing the technical portfolio, and in placing individual research projects into a larger context.
Many of the projects in the networks and communication area use data sets to test and improve the research approaches and solutions. Some data is representative of Army mission scenarios, but a significant fraction of the data being employed is based on academic collections or benign social media data. There is an opportunity to plan and collect data sets with increasingly difficult scenarios, and to integrate these plans into the research campaigns. As an example, data collections could include combinations of humans (friendly and adversary) and machines or robots in appropriately challenging environments.
The innovations in network protocols have the potential for improved operation in diverse and challenging environments. However, past experiences by other DoD agencies have shown significant challenges in transitioning innovative communications and datalinks to the operational Army and coalition forces. ARL has an opportunity to facilitate transition by early planning and diligent management of the transition paths.
There are two potential gaps in the networks and communication research agenda. First, the research presented mostly involves machine-to-machine communications and interactions. Given the significant human-to-human interaction present, and the significant opportunities for human-machine teaming, there is an opportunity to develop more research specifically addressing human elements of communications and interaction, including both human-to-human communications through machines, and human-machine communications and interaction. Second, communication strategies for groups of robots, such as groups of UAVs, were not presented and may be an important enabler for future Army needs.
There are substantial conceptual synergies between the cyber and the network science communications work. The work in both areas could be strengthened by building on these synergies. One example is data fusion for anomaly detection, which appears in both programs. Another synergy is the research on adversarial activity, which could take a combined machine learning and network science approach to move beyond a single adversary cyberattack model.
Event Tracking from Streaming Open Source and Infrastructure Data
This current work is based on public or academic data sets, with limited ground truth information for assessing algorithm correctness. There would be value in testing the approach on more military-relevant data in a contested environment. One possibility is to apply collaboration with Australia and
the United Kingdom on counter-terrorism trials. This is potentially of strong relevance, but there was little planning on how to set up these trials to assess the military relevance of this program. The work could move beyond activity counts, and could consider the changing temporal, spatial, content, and network signature of the data.
Robust Distributed Optimization with Dynamic Event-Triggered Communication
This work could consider performance as a function of network topology, and consider broadening the results to a larger class of optimization or consensus averaging methods. Additional work could also consider what level of consensus is needed to improve performance of soldiers who would use this information. It may also be helpful to consider more tangible demonstration plans aligned with the AI and ML ERA.
Networking in Resource Constrained Environments
This research could consider discussing the applicability of this system with the Special Operations Forces. This continues to be a high-impact program with a test-bed that can be used to prototype and test other innovative research in networking and communications at the ARL.
Quality of Information for Semantically Adaptive Networks
This research project examines an approach for intelligently directing the flow of useful information, employing semantic awareness and information quality together to improve transmission. There is a clear Army need for such tools, but the project description does not explain how the research approach would address this need. The vision presented was to consider all data, regardless of source, and to use contextual cues to pare it down. This builds on prior work on information quality for visual analytics. The current project appears to extend this to the semantic area without addressing the difficulties of assessing context semantically. Future research would benefit from assessing the value of information for each media/data type such as Twitter, Facebook, and imagery, within a common framework. The description was diffuse, and the problem statement would benefit from additional structure and specificity.
Geometric and Topological Structures in Composite Networks
The pathway from this research to more applied and Army-relevant work was less apparent. The topological issues addressed seemed distant from current concerns in network topology. The baseline models used Erdos-Renyi graphs, which are mathematically tractable and elegant; however, most real networks are not Erdos-Renyi, and as a result, the applicability of the results to real networks is unclear. It was unclear how the current focus on properties of stationary networks would address the goal of understanding dynamic networks. Linking the problem statement to topics that have Army relevance and consider topologies aligned to Army operational scenarios would improve the impact of this program.
Decision Making with Uncertain Opinions
This small basic research program is a new start since the previous review. This program considers a model of opinion formation for how an individual in a network changes his or her opinion over time. The research goal is to improve understanding of how people form opinions, so that counter-measures
could be devised to prevent the spread of wrong opinions. It was difficult to extract a crisp problem statement or justification of the modeling assumptions. The model employed is somewhat contrived and not well-connected to both existing social science literature and its well-validated models on opinion formation and social influence theory, or to communication theory models. The SIR model used is not consistent with known human opinion dynamics, and it cannot account for the way cognitive and social biases influence opinion formation. This research program would benefit from a stronger connection to these previously documented studies by building on existing validated models. The work assumes a model where there is a single opinion about an issue and where there is ground truth. This research could be enhanced by considering models in which human opinions are embedded in opinion networks and hierarchies, and in which pro and con opinions simultaneously propagate, and in which there is no ground truth. These modifications would improve the ability to model decision making in Army-relevant scenarios of interest, and thereby improve the Army relevance and potential for transition.
Software-Defined Networking for Tactical Networks
The SDN concepts tested are well-established ideas; future work might consider collaborating with researchers who could pursue new ideas that improve upon results using existing SDN techniques. In addition, the research would benefit from a clear statement of objectives. The definition of SDN used in this research differs from its common use in the literature, which can cause confusion in the presentation of the research; the researchers might consider dropping the use of the term and instead use a different term that emphasizes the self-management characteristic of the networks under consideration.
CYBERSECURITY: DETECTION AND AGILITY
The defense of information systems and networks is a critical need for the U.S. Army. ARL research in the area of cybersecurity seeks to understand and overcome vulnerabilities in this regard, where cyberattackers, both human and intelligent agents, pose a significant and persistent threat. Understanding how adversarial elements interact with information is important, as is the analysis and understanding of adversary resources, learning and recognizing adversary tactics, and ultimately anticipating adversarial activity to mitigate the effects of cyberattacks. Risk characterization is another important element of cybersecurity research, and focuses on metrics and algorithms for risk assessment and risk data collection to support risk assessment.
The cybersecurity area of the Information Sciences Campaign has made notable progress since the last review in 2016. The overall quality of research was good; some projects can be characterized as excellent. ARL researchers engaged in cybersecurity projects were scientifically competent. Given the scarcity of cybersecurity professionals, this is quite noteworthy. There is good evidence of collaborative work with other organizations, particularly in the collaborative research projects detailed in the poster session. More importantly, ARL researchers were generally aware of other research in their areas of work.
Accomplishments and Advancements
The majority of research projects reviewed were relatively new starts and, as such, had little in terms of accomplishments upon which to base an evaluation. All of the research projects reviewed were of merit and appropriate for ARL’s research agenda. The researchers were also knowledgeable about their projects and had the appropriate background and knowledge to succeed.
Adversarial Influence of Machine Learning on Active Cyberdefense
This study started in December of 2017 and is still in its early stages. It seeks to use machine learning techniques that have been largely applied to images in the area of cybersecurity, primarily to find packets in a network that might be infected with malware. Thus far, only a literature review has been conducted. This work is a unique and worthy research topic of particular relevance to the Army.
Active Defense System for Vehicle Platforms
This short-term project is focused on surveying the current state of the art, and building a simulation environment in which to explore prototyping and demonstration of active defense techniques. The approach involves first trying to identify possible vulnerabilities in such systems, a step that the investigator is qualified to do based upon six years of experience in the offense domain of cybersecurity. It was not clear how far the project has advanced to actually prototype promising active defense techniques. The project has a number of partners, both academic and in the U.S. Army, to facilitate transition activities. The development of the prototype and its subsequent assessment would demonstrate whether further research is warranted.
Cyber Experimentation: Data Sets and Software Agents
This research project is focused on building a test-bed for validating AI algorithms using collected data from actual attacks against the network that ARL operates for the DoD. The cyber infrastructure, built by a contractor, is essentially a cyber range. It allows one to simulate a wide variety of end hosts and networking protocols to conduct controlled cyber experiments.
The activities related to this project will take place in the CyberVAN and are more restricted in scope. The focus of this work is on evaluation of tools that will automatically label data. The researchers will generate data sets from CyberVAN, and will manually label these data sets. The labeled data sets will represent ground truth to be used for evaluating the effectiveness of vendor tools that profess to automatically label data sets. This project is a good example of applied research in support of Army operational requirements. The ground truth data set would be of value in other tool evaluations as well, and to the larger technical community.
Stylometry—Authorship Attribution for Source Code and Binaries
In cybersecurity, attribution of perpetrators and weapons is a difficult and elusive goal. There have been numerous research efforts to attribute a cyber weapon to a family of cyber weapons of common origin. Most of these efforts have been aimed at attributing source code to common origin. What makes ARL’s research novel is being able to carry the work over to binaries.
This has been a continuing project for several years and has shown considerable promise. Using this methodology, determining the authorship of existing binaries is fairly reliable. The success of the research thus far has generated several additional research questions. In particular, there is a question of how awareness of the characterization techniques may lead adversaries to filter programs so as to obscure their origin and diminish the value of this technique. New questions have also emerged about how accurate these methods are to community development, binaries produced by automated frameworks, and binaries with huge amounts of useless code introduced to change the natural feature statistics that are leveraged for identification.
Intelligent Rapid Honeynet Generation for Mission-Based Cyberdefense and Resilience Systems
This effort is focused on a game-theoretic approach to the selection of defensive measures involving honeynets (e.g., spoofing nodes and redirecting traffic). Current work is on defining a zero-sum stochastic (Markov) game that represents interplay between an attacker selecting pathways in an attack graph and a defender that has some opportunity to shape the graph. The idea of modeling adversarial behavior through games is laudable. The ongoing approach is ground-up, in the sense that it considers abstract models that involve a number of simplifying assumptions (e.g., zero-sum rewards) and then computes optimal strategies for the players.
Malware and Intrusion Detection for Mobile Ad Hoc Networks (MANETs)
This use of Bloom filters as a mechanism for compressing a large corpus of signatures in a network-constrained environment is ingenious. The technique hashes, using three hash functions (although this could be easily adjusted), n-grams (currently, 3 bytes) of the packets containing known malware to a Bloom filter (currently containing 224 bits), and setting the hashed-to bit to 1. Subsequently, packets whose contents are unknown have their 3-grams hashed similarly, and if all bits hashed to are 1, that packet almost certainly (modulo very unlikely hash collisions) contain that same malware. Preliminary testing of the approach used with malware packets against known data sets has shown an order of magnitude bandwidth savings in mobile tactical networks with false positive rates of less than 6 percent. This project is an excellent example of using ARL research to solve a problem that is pertinent to the Army warfighter environment. It illustrates the successful adaptation of a solution strategy from one problem domain into another.
Cyber-CAMO: Mission-Based Cyberdefense and Resilience for Tactical Networks
This research project seeks an approach to active defense by replicating a number of nodes in order to avoid single points of failure. The approach mixes these nodes to create different channels that now connect to a mixed cloud of nodes. These diverse clouds of nodes create a type of game theoretic situation in which an attacker must choose which connection to jam. The game theoretic formulation allows an approach for predicting how the attacker will act and how the defender should respond.
Modernizing the Army Cyber-Research Analytics Capability
The Army cyber-research analytics capability (ACAL) environment is both a formalized process and a hardware environment to support storing and performing data analytics on large collections of cyber data. The framework includes methods for cleaning, fusing, and storing the data, as well as application programming interfaces for user interaction. ACAL is a critical component in ARL’s efforts to allow researchers to learn from the very large amount of cyber data that the laboratory collects on a routine basis. ACAL is organized to support general data analytics on cyber data, and is not tailored to adversarial frameworks, scenario simulations, or other applications that might support direct testing of ML and AI methods being developed in the other projects.
This work in this project is developing a process for approving access to ARL operational data for collaborators outside ARL, including other government agencies.
The project is a good example of ARL’s continued efforts to optimize the utilization of its unique cyberattack and defense data.
Comparative Assessment of a Packet Reduction Method to Preserve Alert Frequencies
This project uses semantically driven sampling to reduce the volume of data collected for cyber-relevant characterization. This is a fairly mature project that has demonstrated its usefulness on real hardware. Past results from the work have shown that the approach preserves 99.3% of indicator signatures using only 10 packets per session. This is an excellent project that transcends idea and theory into implementation and practice. The researchers need to focus on developing and pursuing transition partners to ensure that the value of the work is retained when the project terminates.
Model Propagation Paths and Characteristic Features of Deceptive Data Campaigns
This project entails very encouraging work that is in keeping with accepted standards in the field. It uses linguistic analysis of the content in fake news web pages known to originate from foreign sources (e.g., Russia) to identify characteristics that will enable automatic identification of other such fake news web pages. While the awkward English constructs in sentences of these pages are usually evident to a human, automating such recognition is challenging. This relatively new study has already revealed a rather surprising result—that the English of the fake news sites contains a higher occurrence of adjectives and adverbs. The goal of this work is to rapidly identify and take down fake news sites before they can be widely viewed or replicated.
ARL CSSP: Data—Research—Development—Operations
This research examines the database of real attacks against the operational network operated by ARL for the Army and other DoD agencies to study how to sufficiently sanitize these data so that they can be externalized. The investigation is not sufficiently mature to allow such release of any data as yet. The real-world incidents gleaned from this operational data constitutes an extremely valuable resource unique to ARL that could be used to validate prototypes in other ARL studies.
Business Intelligence for Cyber Operations
This project researches the deployment of business intelligence tools on operational data gleaned from ARL’s cybersecurity network, utilizing off-the-shelf software (Tableau) to develop dashboards, rather than generating them using slideware (e.g., PowerPoint) from raw statistics, thereby reducing the manual effort and time to generate meaningful analytics on the occurrence of incidents. While not defined as fundamental research, the work does accommodate an important need of Army cyber operators and their commanders.
Challenges and Opportunities
ARL’s position and history have created unique opportunities for research. As has been noted in previous reviews, the greatest resource ARL has in cybersecurity research is its continuing access to real-world data and real-world operators. ARL needs to utilize these resources to the largest extent possible in its cybersecurity research programs, as they provide nearly the only ground truth in research related to cyberattacks. It is encouraging to see research specifically focused on providing greater access to that data, both internally and externally.
Research related to the security of cyber-physical and mobile systems needs to be a priority of ARL. Many of the cyber-physical and mobile systems in the Army are unique to the Army, and no one else is likely to expend research effort in this arena. The security of these systems is of critical importance to the Army and the nation at large. In particular, the safety and security of battlefield autonomous vehicles is of importance for the Army, and this an opportunity for the ARL to be a research leader in the area. ARL also faces challenges, including balancing long-term research objectives with short-term mission needs. This also includes balancing a leading-edge fundamental research agenda against more applied solutions addressing Army-specific mission requirements. As ARL shifts to more long-term research projects, it will need to establish interim objectives and goals so that direction and progress can be more easily assessed. The cybersecurity research staff at ARL seems well-equipped to contribute in a meaningful manner to the projects. However, high demand, coupled to scarcity of cybersecurity professionals, will likely challenge ARL to retain and recruit the best and brightest professionals in this field of work.
Adversarial Influence of Machine Learning on Active Cyberdefense
Adversarial ML is an important area, as recent work in the broad ML community has demonstrated. Image recognition classifiers and other types of models can be tricked into providing incorrect answers if small amounts of training or input data are altered in precisely calculated ways. This project involves limited work that would demonstrate similar attacks in a cyber environment. The bulk of the effort is directed at identifying defense mechanisms. The project requires the construction of a simulation environment that would allow testing of defensive algorithms against attack algorithms in a variety of scenarios.
The researchers do not appear to have extensive experience in AI or ML, but they are asking interesting questions. Their prior experience in understanding the adversary’s perspective will be of immense value to the research.
Active Defense System for Vehicle Platforms
Active defense techniques are of increasing importance for electronically controlled and autonomous vehicles. The lack of a centralized bus control, lack of access control and message authentication, and need for real-time response in such vehicles add considerable complexity and challenge to this problem.
Stylometry—Authorship Attribution for Source Code and Binaries
The researchers of this project need to collaborate with virus hunters in the commercial sphere, and routinely attend black hat conferences to become familiar with the best techniques employed for identifying and chasing down malware authors.
Game-Theoretical Defensive Deception Techniques for Dynamic Honeynet
This new project proposes the use of game theory concepts to develop a taxonomy of techniques for defensive deception. Specifically, it addresses the interesting and important problem of using deception to dynamically generate honeynets. The project seeks approaches that would rapidly characterize attackers to determine the best method for deception and honeynet creation.
Although this work addresses an interesting problem, it is not clear where or why the game theoretic approach would be deployed. Even though defensive deception has not been thoroughly studied, it may
be useful for the researchers to consider a data- or case-based approach to characterizing attackers and developing defensive strategy. The project needs to more tightly define how it will approach this problem and what the metrics for success relative to cost are.
Intelligent Rapid Honeynet Generation for Mission-Based Cyberdefense and Resilience Systems
The proposed approach presumes that the attacker cannot use a foothold within the network to undermine the inputs to the game-theoretic model, thereby compromising the defender’s optimal strategy. As an example, the attacker could gain access to a poorly defended unimportant node (e.g., a minor sensor) that is vulnerable, and then spoof the node to be highly important and highly connected (e.g., the node of the commanding officer). This would deeply compromise the otherwise excellent model. Also, it may be appropriate to consider value and vulnerability as independent metrics of each node and the connection between each node (assuming it is not a fully connected network). The value is a function of the importance of that node (e.g., commanding officer versus a minor sensor), whereas the vulnerability is a function of the hardware and software associated with that node, and how well it is physically defended. In this scenario, the game-theoretic model would incorporate multiple metrics through a multicriteria optimization approach.
Researchers in this project could also consider approaching the issue from the converse perspective by identifying sets of actions and ultimately classes of strategies that could be implemented in the battlefield. This perspective would involve a number of nontrivial issues, such as what a randomized strategy might imply in terms of doctrine and warfighter training. A model or framework of this type would support testing defense heuristics based on insights from game theory at large.
Cyber-CAMO: Mission-Based Cyberdefense and Resilience for Tactical Networks
As conceived, the approach requires additional clarification and validation. At the very outset, the strategy alternatives in a realistic attack scenario are not self-evident. Similarly, the advantages of the game-theoretic approach are not apparent given the reliance options (duplication of some processes). For example, if the diversity introduced to achieve mission objectives is limited to a choice of one of five RF transmission channels, what is left for the game theory if all of the channels are defeated? An analysis of concrete gains and losses, as well as the dynamics of the choice alternatives at a game node would be useful.
Replicating data across nodes to increase survivability is a well-known technique in commercial data networks that has been extensively studied for decades. The unique aspects of this work beyond the intentional introduction of diverse hardware and software, and the frequency with which nodes fail, need to be articulated.
Model Propagation Paths and Characteristic Features of Deceptive Data Campaigns
This work is at an early stage, and determining how robust these techniques will be against an informed adversary is an interesting future question. It would also be important to think about possible actionable responses as well as how these detection techniques can be leveraged to develop a reaction.
OVERALL QUALITY OF THE WORK
The research portfolio in the Information Sciences Campaign reviewed in this current 2-year cycle was expansive, covering areas spanning sensing and effecting, system intelligence and intelligent systems, human information interaction, atmospheric sciences, networks and communications, and cybersecurity. The projects reviewed range from those advancing fundamental science to those focused on enabling technologies and applications. The ongoing projects demonstrated relevance to the future missions of the Army and were generally of good technical quality. There are additional opportunities to further drive scientific innovation through enhanced integration and collaboration across campaigns.
ARL has focused on increasing the number of Ph.D. scientists on the research staff in critical areas of expertise, and this has had a positive impact on the overall quality of technical work. The nature of the research in information sciences dictates a workforce with significant technical diversity, including strengths in the social and mathematical sciences. This diversity and added strength in the social sciences is also critical to the HII initiative. Among the ARL researchers, there was generally a good awareness of external research and connections to professional organizations and external research communities; research results are appearing in respected conference proceedings and in archival journals. There is room for even broader dissemination of these results to a larger scientific community. As noted in earlier reviews, the mission-oriented thrust helps differentiate the ongoing research from efforts pursued elsewhere, and it creates opportunities for impactful technical contributions. The impact of the work can be further enhanced by clear articulation of unique, cutting-edge research questions.
The work in sensing and effecting was generally of high scientific research quality, but not uniformly so. The research portfolio represents an appropriate balance of theoretical and experimental work, and many of the more mature programs, but not all, show a transition into practice or use by other areas. The research generally reflected a good understanding of the problems being considered, appropriate statement of the problem being pursued, knowledge of the appropriate methodologies to address the problem, and a good knowledge of the state of the art and the relevant research pursued elsewhere. In many cases, the researchers are able to articulate Army relevance and unique aspects to Army needs when that was the case. The researchers are academically well qualified to carry out the research problems that they are pursuing. Some of the work is being published in top venues, but researchers could be given additional guidance and encouragement to present their work at leading conferences, and then pursue publication in top archival journals. The increase in postdoctoral researchers and early-career researchers is a positive trend. The quality of the early-career ARL researchers in the S&E area is impressive. The ARL process of using co-ops and internships to recruit a strong cadre of emerging researchers appears to be yielding positive results. The facilities are adequate to the needs of the researchers and of the research, including access to computational facilities and to laboratory instrumentation.
The overall research program in SIIS is aligned with the current and future needs of the Army. It is clear that the researchers within SIIS are generally conversant with modern techniques and approaches from the rapid advances in machine-learning research. SIIS researchers were forthcoming and articulate in their presentations and responses to questions. Thorough materials were provided well in advance of the meeting, newer publications were provided, and the research team was available for clarification after its initial presentations. The presentations of the different projects were of variable quality. In part, this was due to the limited scope of some of the projects, or that the project was in an early stage and did not have a comprehensive set of results. In some instances, the presentations simply failed to highlight key questions, methods, and interim or final results. For some of the more experimental work, it was not clear whether the researchers were building an engineering artifact to develop initial insights and validate approaches, or whether they had a more scientific paradigm that started with a hypothesis that
could be crisply stated and that was the basis for the design of the experiment. In many instances, the research results are being disseminated in top conferences or journals, but not uniformly so. While not every piece of good work will be published in the most highly regarded venues, the site of publication is a good indication of high-quality work and provides excellent visibility for the researcher and ARL. Furthermore, publishing in high-level journals will attract the top academic researchers and their students to problems most relevant to the ARL mission.
The research focus in the area of HII is relatively new and includes many projects that are at an early stage of inception. As a result, the technical quality of the work is mixed. The projects were highly variable in the extent to which they had identified the relevant computational and social theories. For the nascent research projects, there is the need to develop a stronger Army research focus and to use this focus to narrow the scope of work as a path to realizing meaningful results. The more mature work has strong Army relevance but tends to focus on single human-technology interaction. While important for demonstration purposes, the impact of the research could be enhanced by reformulating the problems to include scenarios involving Army teams interacting with information and technology. The researchers are skilled. Early-career researchers could use more mentoring, and all researchers would benefit from better ties to the broader scientific community. The field is highly interdisciplinary, and research productivity and impact would be enhanced if the HII team was better balanced to include additional social scientists. The work being done by HII researchers is being disseminated at relevant conferences, including multiple papers at leading venues such as the SBP-BRiMS conference.
In work related to atmospheric sciences, the overall scientific quality of the work is very good, and comparable (and, in a few cases superior) to research conducted at successful university, government, and industry laboratories. Researchers are very familiar with the underlying science and cognizant of research being done elsewhere; in many cases, they have been either dialoguing or collaborating with researchers outside ARL. In all cases, the researchers are aware of the potential challenges associated with their projects. BED uses the use-inspired research4 approach, resulting in researchers from different disciplines being brought into their projects. In one case, the dialogue led to the recent staff additions of a mathematician from another ARL division and a NASA-supported scientist, which have greatly benefited the vortex filament and lattice Boltzmann projects. In several other instances, collaboration and consultation with researchers outside atmospheric sciences have enhanced project quality and applicability to the Army’s mission. These types of approach are commendable and need to be continued by the BED leadership.
In the area of network and communications, the overall scientific quality of the work is very good, and quite comparable (and, in a few cases superior) to research conducted at successful university, government, and industry labs. Researchers are in most cases familiar with the underlying science and cognizant of research being done elsewhere. In many cases, they have been either communicating or collaborating with researchers outside ARL, many at the forefront of their respective fields. In nearly all cases, the researchers are aware of the potential challenges associated with their projects. In many cases, the researchers were able to incorporate these challenges into research strategies that address them, but this was not always the case.
In the cybersecurity area, most projects reviewed were relatively new, with few results upon which to base a detailed evaluation. That said, the quality of the work was high and relevant to ARL’s mission. In general, the work is on a par with that at other government laboratories, and in some areas
___________________
4 A commentary on use-inspired basic research, “Thinking Beyond the ‘Quadrant,’” by Dr. F. Fleming Crim, NSF Assistant Director for Mathematical and Physical Sciences, https://www.nsf.gov/mps/perspectives/quadrant_august2014.jsp (accessed August 10, 2017).
ahead of research at academic institutions. This is particularly true in some areas of network defense, where ARL has a huge advantage in its access to a facility that provides actual data. Researchers at ARL are less isolated from peers than noted in previous reviews, are generally aware of related work done at other institutions, and are actively participating in conferences and workshops. Opportunities for outside cooperation, such as the extramural research and open campus, have had a positive impact and are further encouraged. The overall quality of the work has improved, but some of the posters and presentations could be further enhanced. In particular, a succinct and clear statement of the research problem needs to be included in all presentations. Also, the specific advantages of a solution strategy to other competing approaches for a problem domain could be included in the presentation. This was not clear in some of the reviewed work.
CONCLUSIONS AND RECOMMENDATIONS
The ongoing projects showed a mix of fundamental and applied research with relevance to the future missions of the Army and were generally of good technical quality. There are additional opportunities to further drive scientific innovation through enhanced integration and collaboration across campaigns.
The work in S&E was generally of high scientific quality, but not uniformly so. There was a balance between theoretical and experimental work, and in many of the more mature programs, there was evidence of transition into practice or use by other areas. The research generally reflected a good understanding of the problems being considered. The researchers are academically well qualified to carry out the research problems that they are pursuing, and some of the work is being published in top venues; more, however, can be done in terms of mentoring and encouragement to seek the most highly rated venues for disseminating their results. The quality of the early-career ARL researchers in the S&E area is impressive, and the capacity to successfully address much more challenging research problems now exists within ARL. There is currently a mismatch between the quality of the researchers and the degree of difficulty of problems that are being pursued.
Recommendation: ARL should actively encourage researchers to take on research problems of greater complexity and scope.
The research in SIIS is appropriately aligned with the emergent priority accorded to the fields of artificial intelligence and machine learning, and is seeking to address key gaps relevant to the Army. Several early-career researchers have been recruited and have the appropriate background and academic training to pursue the research goals that have been identified. The research projects are appropriately resourced, and both the facilities and the computational resources are adequate for the proposed work. There is a healthy collaboration among researchers, both within ARL and with the external community. Continued investments in this regard would be worthwhile, because in some instances it appeared that SIIS researchers within ARL were not always aware of similar or parallel efforts elsewhere.
Recommendation: ARL should bring about greater understanding of available technical expertise in system intelligence and intelligent systems-related areas across ARL, and should build synergy across campaign thrusts to leverage this technical talent.
HII is a small program that seeks to address an enormous area, with a wide range of potential applications. HII would be well served by developing more within-group synergy and concentrating on those aspects that are highly relevant to Army missions and the deployed dismounted soldier. Given the
nascent nature of the program, it is advisable to develop a strong mentoring program, perhaps using a combination of external and internal mentors. There is also a need for developing a broader set of external collaborators, including mathematical sociologists and social psychologists in the area of teaming. In this context, it would be helpful to send scientists to leading conferences in the area and to organize internal seminars and workshops in advanced areas such as ABMs, information diffusion and the spread of misinformation, and social media analytics. The work in HII is hampered by not having access to all state-of-the-art tools and software. This concern could be alleviated by making tools such as Repast and ORA and various image processing software available to researchers.
Recommendation: ARL should develop a strong mentoring program for the research team involving external and internal mentors and should establish a broader set of external collaborations to accelerate progress of the research agenda.
With respect to ongoing research in BED, the importance of the natural environment needs to be featured more prominently in strategic planning documents in order to advance the fundamental science while encouraging collaboration and dialogue between researchers whose projects have an environmental sensitivity and subject matter experts (SMEs) from BED. The exclusion of environmental effects, whether inadvertent or intentional, could possibly jeopardize the successful fielding of promising new technologies to the Army and broader society. Environmental phenomena are crosscutting and impact nearly every Army operation.
Recommendation: The environmental impact on Army operations should be given crosscutting, prominent visibility in planning documents such as the science and technology campaign plan and the essential research areas that are being formalized by ARL senior leadership. ARL should evaluate the potential for environmental integration at the initiation of a new project, to enable teams to incorporate multiple components into their approach at the inception of the design process, and potentially enable faster basic science and applied technology development.
The networking and communication area continues to address basic and applied research to advance technology related to the understanding of dynamical behavior of networks and the interactions between information and social networks. The research emphasizes development of technologies that apply to unconventional networks that will employ heterogeneous approaches to encoding and transmitting information, with a focus on robustness and survivability in harsh operating environments.
Recommendation: ARL should continue to plan and collect data sets with increasingly difficult scenarios, and should integrate these plans into the research campaigns. Data collections should include combinations of humans (friendly and adversary) and machines or robots in appropriately challenging environments. ARL should also broaden the research agenda to more fully address human elements of communication and interaction, including both combined human-machine communication and information understanding.
In the cybersecurity area, most of the projects were relatively new, and the limited results did not provide a firm basis for detailed evaluation. That said, the research directions presented were noteworthy and would be welcome in any cybersecurity research laboratory. The dual nature of the cybersecurity role that ARL plays—operations and research—creates a unique capability for research in this field. ARL is one of the few cybersecurity research organizations with continuing access to real-world data—the
importance of which cannot be overstated. Additionally, ARL researchers are interfacing with a deployed operational environment. This gives ARL an ability to focus on research that has an immediate bearing and impact. All of the research projects reviewed showed suitable scientific rigor and practice. A wide variety of methodologies and scientific approaches were included in the reviewed projects.
Recommendation: To provide a clearly articulated vision of a technical area in which ARL could lead research, ARL should set one or more research challenge goals and use those to focus the research portfolio. One possible challenge goal that would be of interest to the Army and the wider defense community and to the commercial world, and that ARL should therefore consider, is the security of autonomous vehicles in a contested environment.
