Given that many readily available datasets consist of large amounts of unlabeled data,1 unsupervised learning methods are an important component of many data-driven applications. In many instances, ground-state truth labels may be unavailable or obtainable only at a costly expense. As a result, there is an acute need for the ability to understand and interpret unlabeled datasets as thoroughly as possible. In this article, we examine the effectiveness of learned deep embeddings via internal clustering metrics on a dataset comprised of unlabelled StarCraft 2 game replays. The results of this work indicate that the use of deep embeddings provides a promising basis for clustering and interpreting player behavior in complex game domains.

An important topic in cybersecurity is validating Active Indicators (AI), which are stimuli that can be implemented in systems to trigger responses from individuals who might or might not be Insider Threats (ITs). The way in which a person responds to the AI is being validated for identifying a potential threat and a non-threat. In order to execute this validation process, it is important to create a paradigm that allows manipulation of AIs for measuring response. The scenarios are posed in a manner that require participants to be situationally aware that they are being monitored and have to act deceptively. In particular, manipulations in the environment should no differences between conditions relative to immersion and ease of use, but the narrative should be the driving force behind non-deceptive and IT responses. The success of the narrative and the simulation environment to induce such behaviors is determined by immersion, usability, and stress response questionnaires, and performance. Initial results of the feasibility to use a narrative reliant upon situation awareness of monitoring and evasion are discussed.

Insider Threats (ITs) are hard to identify because of their knowledge of the organization and motivation to avoid detection. One approach to detecting ITs utilizes Active Indicators (AI), stimuli that elicit a characteristic response from the insider. The present research implemented this approach within a simulation of financial investigative work. A sequence of AIs associated with accessing a locked file was introduced into an ongoing workflow. Participants allocated to an insider role accessed the file illicitly. Eye tracking metrics were used to differentiate insiders and control participants performing legitimate role. Data suggested that ITs may show responses suggestive of strategic concealment of interest and emotional stress. Such findings may provide the basis for a cognitive engineering approach to IT detection.

Many cyber security officers are more concerned with outside rather than insider threats because the enemy is generally perceived as being “out there” or beyond the organization. Therefore, defensive actions are readily available once an outside threat is identified (Colwill in Human factors in information security: the insider threat—who can you trust these days? pp. 186–196, 2009 [1]). Contradictory to the ideas of social identification as an “us” and “them,” the greatest enemy may be lurking within one’s own organization. Individuals are considered insiders if they presently have (or at one time had) permission to access an organization’s data or network structures (Greitzer et al. in Secur Priv IEEE 6(1):61–64, 2008 [2]). The concept of the insider threat is considered one of the most difficult situations to deal with in the cybersecurity domain (Hunker and Probst in J Wireless Mobile Netw Ubiquitous Comput Dependable Appl 2(1):4–27, 2011 [3]). The Association of Certified Fraud Examiners has reported two-thirds of fraud and identity thefts are executed by organizations’ employees or other known insiders. They also estimate U.S. companies have lost 5 % of revenue to fraudulent insider activities (Randazzo et al. in Insider threat study: illicit cyber activity in the banking and finance sector, 2005 [4]). Insiders have multiple advantages over an outsider. An insider threat is one of the most difficult situations to identify. Therefore, it is critical that training be developed. The first step to effective training is constructing an environment that lends itself to insider threat situations. The present paper describes the process in which one insider threat virtual environment was constructed. A discussion of the considerations and functional features is detailed.

The United States Army has heavily leveraged, developed and expanded its use of virtual simulation training, as this class of simulation has been empirically demonstrated to be effective in the transfer of skills to the live environment. Game-based training, an alternative class of simulation, is characterized by its lower overhead and cost and potentially represents a less expensive alternative to virtual simulation training. In an effort to reduce the cost of training simulation, the U.S. Army has recently socialized the concept of potentially replacing select virtual simulation trainers with game-based simulations. While lowering the cost of simulation is a noble endeavor, the aforementioned concept requires further investigation as minimal empirical evidence exists regarding the effectiveness of game-based training, particularly at the collective echelon of training. In this paper, we lay the foundation to conduct an investigation of whether a game-based simulation may be as equally an effective collective training apparatus as a higher-cost, higher-fidelity, virtual simulation. Specifically, we discuss a planned Training Effectiveness Evaluation (TEE) of both the Aviation Combined Arms Tactical Trainer (AVCATT) and a game-based aviation simulation that will empirically determine whether or not virtual training in the AVCATT could potentially be replaced by an equally effective, but less costly, game-based simulation. We discourse on our proposed design of experiment, which will utilize qualified Army aviators performing a tactical, collective mission in two discrete training treatments (AVCATT and a game-based simulation) at Fort Rucker, Alabama.

The ability to read the human terrain is an invaluable skill set developed by Warfighters to identify irregular behavior in any environment. Traditional training methods (i.e., classroom-based instruction) have evolved to include SimulationBased Training (SBT). SBT allows virtual representation of behavioral cues for enhancing combat training skills. Accurately modeling human behavior cues in a Virtual Environment (VE) is critical for success. Past research has highlighted difficulties with visually representing proxemics behavioral cues in a VE, however recent efforts have been made to offer insight into design requirements. Proxemics involves the spatial distancing between individuals given the climate of the situation and environment. This research seeks to identify design requirements and recommendations for representing proxemic cues within a VE. Specifically, this paper focuses on the development process of identifying, designing, and representing virtual models that exhibit proxemic cues. Finally, the paper discusses limitations and future directions for behavior cue training.

Computer animations exhibit the illusion of movements or actions of virtual agents and assets within a virtual environment display. Two distinct animation categories exist: two-dimensional (2D) and three-dimensional (3D). 2D animation is typically stylized and used primarily for entertainment-based efforts such as cartoons and lowfidelity games. 3D animation is applied to a wider variety of domains (e.g., entertainment games, serious games, and training simulations). A well-designed 3D computer animation enables a realistic representation of action portraying the true context of movement, particularly human gestures (Badler, Palmer, & Bindiganavale, 1999). All humans convey intent whether purposefully or not via verbal and non-verbal cues (Bavelas, 1990; Givens, 2002). Kinesic cues convey information to an observer through body language and gestures. Emerging research in training human threat detection requires virtual agents exhibiting kinesic cues to provide visual stimuli within Simulation-Based Training (SBT) applications. Thus, guidelines and specifications for system developers are required. This paper presents a process for defining, designing, and animating kinesic cues using a commercially available software application to mimic realistic human behaviors, movements, and gestures. Through this discussion, culturally agnostic kinesic cues are presented, and relevant limitations are identified. The process described and lessons learned represent a logical progression in the formalization of developing advanced visual models for training Warfighters, law enforcement agents, and first responders to detect and classify human threats.

Cybersecurity is a computer term regarding the detection, anticipation, and prevention of computer technologies and peripherals from damage, attack, or unauthorized access. These technologies include the monitoring of networks, programs, applications, and personnel. Cybersecurity can be viewed from both an offensive or defensive posture involving maintaining and proactively assessing security vulnerabilities. In 2013, Edward Snowden used his position as an infrastructure analyst to leak thousands of top-secret classified documents detailing the U.S. Government’s global covert surveillance and eavesdropping undertakings to the public. This incident identified the human threat as a contributing factor that highlighted several weaknesses in the present state of U.S. cybersecurity affairs. In efforts to strengthen cyber defenses, a solid theoretical research foundation regarding cyber vulnerabilities is warranted. Building upon that foundation, training and experimentation can provide insight into current cybersecurity training methods and how they can be transitioned and implemented into future training regimens.

Combat Profiling techniques strengthen a Warfighter’s ability to quickly react to situations within the operational environment based upon observable behavioral identifiers. One significant domain-specific skill researched is kinesics, or the study of body language. A Warfighter’s ability to distinguish kinesic cues can greatly aid in the detection of possible threatening activities or individuals with harmful intent. This paper describes a research effort assessing the effectiveness of kinesic cue depiction within Simulation-Based Training (SBT) systems and the impact of engagement levels upon trainee performance. For this experiment, live training content served as the foundation for scenarios generated using Bohemia Interactive’s Virtual Battlespace 2 (VBS2). Training content was presented on a standard desktop computer or within a physically immersive Virtual Environment (VE). Results suggest that the utilization of a highly immersive VE is not critical to achieve optimal performance during familiarization training of kinesic cue detection. While there was not a significant difference in engagement between conditions, the data showed evidence to suggest decreased levels of engagement by participants using the immersive VE. Further analysis revealed that temporal dissociation, which was significantly lower in the immersive VE condition, was a predictor of simulation engagement. In one respect, this indicates that standard desktop systems are suited for transitioning existing kinesic familiarization training content from the classroom to a personal computer. However, interpretation of the results requires operational context that suggests the capabilities of high-fidelity immersive VEs are not fully utilized by existing training methodologies. Thus, this research serves as an illustration of technology advancements compelling the SBT community to evolve training methods in order to fully benefit from emerging technologies.

Combat Profiling involves observation of humans and the environment to identify behavioral anomalies signifying the presence of a potential threat. Desires to expand accessibility to Combat Profiling training motivate the training community to investigate Virtual Environments (VEs). VE design recommendations will benefit efforts to translate Combat Profiling training methods to virtual platforms. Visual aspects of virtual environments may significantly impact observational and perceptual training objectives. This experiment compared the effects of high and low fidelity virtual characters for biometric cue detection training on participant performance and perceptions. Results suggest that high fidelity virtual characters promote positive training perceptions and self-efficacy, but do not significantly impact overall performance.

The increasing demand of unmanned systems, such as Remote Weapon Systems, for the U. S. military requires expedited empirical research investigating the dynamics associated with these systems and their interaction with Soldiers. Limited availability and accessibility of active duty Soldiers for study participation presents challenges for researchers to conduct relevant military-related research. The purpose of this study is to propose the use of non-military novices, such as university students, as participants for initial hypothesis testing and pilot experimentation in Remote Weapon Systems research. Previous experiments conducted with active duty Soldiers were replicated with university students to compare patterns of performance and cognitive load. Results indicate several similarities between Student and Soldier performance patterns, thus, justifying the use of non-military novices for research with Remote Weapon Systems.

In the United States Military, 2011 marks the third year of a 25 year plan to increase the number of unmanned systems across the air, ground, and maritime domains. These systems perform as members of human-robot teams either autonomously or by remote-operation. The success of employing unmanned systems in coordination with human team members depends on system capabilities which support teamwork and trust. Weaponization of these systems introduces new concerns in teamwork and trust research. This paper presents research comparing the effects of autonomous and remotely-operated unmanned weapon systems on human-robot teamwork and trust. The results will contribute to the development of recommended roles and automation levels for future weaponized robotic systems.

The U.S. military integrates unmanned systems within combat operations with greater regularity and scope each year. Warfighters currently conduct operations such as Improvised Explosive Device (IED) interrogation, and Unmanned Aerial System (UAS) surveillance and reconnaissance with various unmanned systems. Integration of unmanned weapon systems into human Fire Teams represents the next evolution in mixed-initiative teams. Such integration aims to improve individual and team performance; however, improved understanding and application of Human-Robot Interaction (HRI) principles within combat environments is required. The research presented investigates the impact upon individual and team performance when a non-autonomous unmanned weapon system is integrated into a human Fire Team. Studies were conducted at two U.S. Army installations involving 36, four-person Fire Teams. At the first location, participants included pre-deployed novice soldiers, and the second installment included experienced soldiers as participants. All soldiers had previous weapon experience using an M16 rifle and M240B machine gun. Two conditions were compared: Fire Teams fully manned by human Warfighters and Fire Teams where one human Warfighter was replaced by a remotely operated weapon. Each team consisted of four members: one M240B Gunner and three M16 Rifleman. The teams completed simulated missions utilizing the Engagement Skills Trainer 2000 (EST 2000), a virtual training simulator that executes various mission scenarios. The Fire Teams completed two different scenarios, each consisting of a manned and unmanned condition. In the unmanned condition, the Gunner operated a remotely operated weapon from a separate location. Performance was primarily measured by recording the total number of targets hit by each team member during scenario execution. Paired samples t-tests revealed significant differences in individual performance from the manned to unmanned conditions. Individual Riflemen improved performance from manned to unmanned scenarios. However, the Gunners significantly decreased in performance when operating the remote weapon system during the unmanned condition. Team performance did not reveal a significant difference in performance across conditions. This paper describes the experimental plan and methodology, followed by a discussion of experimental results and recommendations for future mixed-initiative team research.