This PDF file contains the front matter associated with SPIE Proceedings Volume 12117, including the Title Page, Copyright information, Table of Contents and Conference Committee list.

Poisoning attacks on training data are becoming one of the top concerns among users of machine learning systems. The goal of such attacks is to inject a small set of maliciously mislabeled training data into the training pipeline so as to detrimentally impact a machine learning model trained on such data. Constructing such attacks for cyber applications is especially challenging due to their realizability constraints. Furthermore, poisoning mitigation techniques for such applications are also not well understood. This paper investigates techniques for realizable data poisoning availability attacks (using several cyber applications), in which an attacker can insert a set of poisoned samples at the training time with the goal of degrading the accuracy of the deployed model. We design a white-box, realizable poisoning attack that degraded the original model’s accuracy by generating mislabeled samples in close vicinity of a selected subset of training points. We investigate this strategy and its modifications for key classifier architectures and provide specific implications for each of them. The paper also proposes a novel data cleaning method as a defense against such poisoning attacks. Our defense includes a diversified ensemble of classifiers, each trained on a different subset of the training set. We use the disagreement of the classifiers’ predictions as a decision whether to keep a given sample in the training dataset or remove it. The results demonstrate the efficiency of this strategy with very limited performance penalty.

One needs a good communication cost model [1, 2] for optimizing the off-loaded computation in a tactical environment. Recently, we presented a mathematical cost model protocol for optimizing those computations [3]. It applies to the Autonomous Mobile Agents (AMA) in the field communicating via a resource-constrained multi-node tactical network. In the present work, we will include the delay and recast the situation as a Linear Programming optimization problem.

Chaos Engineering (CE), which Netflix introduced in 2008, is used by researchers to assess and find weaknesses in system resiliency. Such weaknesses can arise, when subsystems are individually robust, but that robustness disappears when multiple subsystems are paired together in a System of Systems (SoS). CE researchers develops methods and metrics for finding such fragilities. In this paper, we expand previous examinations of CE experimentation for SoS and introduce Security Chaos Engineering (SCE) for SoS. These SCE experiments include terminating message service, flooding multi queues/message, and injecting corrupted Service. SCE assumes compromise by adding a malicious actor to the tests that can induce adversarial failures into a SoS. For our SoS testbed, we instantiated a virtual Unmanned Aerial Vehicle (VUAV). We use the open-source Chaos Toolkit to run consistent CE and SCE experiments on the VUAV. Chaos Toolkit with SCE exposes the VUAV attack surfaces to evaluate performance and system security. This research allows us to establish an understanding of baseline system performance and gaps in procedures, techniques, and tools from the state of the art as applied to DoD-relevant systems like SoS. We use the load placed on the Central Processing Unit (CPU) and Random-Access Memory (RAM) by the VUAV as metrics for baseline performance. The results showed that these two metrics did not provide enough fidelity in where CE/SCE creates failures. Feeding these results into the CE methodology allows for additional metrics to better pinpoint failures with CE/SCE testing.

White Cell adjudication evaluates impact of player moves to create randomness representing realistic scenarios. Players are limited to the view of the game state based on their technologies’ capabilities and intel sources, while the White Cell knows game state truth. The Automated White Cell (TAWC) is developed to provide adjudication guidance as to how much and what types of testing are sufficient to determine acceptable levels of randomness to evaluate AI/Autonomous platforms, including manned-unmanned teaming scenarios. TAWC is based on multi-model data integration, using AI/ML/Game Theory solutions, enhancing the predictive fidelity of Test and Evaluation, Verification and Validation (TEVV) /Live Virtual Construct (LVC) facilities in assessing AI/Autonomous technology strengths and weaknesses.

Many believe that humans need to build trust in the artificial intelligence that they use or collaborate with. Transparency and accountability are two fundamental requirements for building that trust. In this paper, we argue that tracking decision provenance is a foundational capability for providing transparency and accountability. The provenance model used in the research described is a simple one, standardized by the World Wide Web Consortium. Provided are descriptions of research that aim to discern critical information about decisions made by autonomous agents through the graphs built by tracking provenance, despite the simplicity of the model and the possible granularity of the resulting graph. The use of provenance to provide explanations of decisions is also described, utilizing Rhetorical Structure Graphs to add application domain and presentation domain knowledge to the spare provenance data model.

A critical factor in utilizing agents with Artificial Intelligence (AI) is their robustness to novelty. AI agents include models that are either engineered or trained. Engineered models include knowledge of those aspects of the environment that are known and considered important by the engineers. Learned models form embeddings of aspects of the environment based on connections made through the training data. In operation, however, a rich environment is likely to present challenges not seen in training sets or accounted for in engineered models. Worse still, adversarial environments are subject to change by opponents. A program at the Defense Advanced Research Project Agency (DARPA) seeks to develop the science necessary to develop and evaluate agents that are robust to novelty. This capability will be required, before AI has the role envisioned within mission critical environments.

Malware is a term that refers to any malicious software used to harm or exploit a device, service, or network. The presence of malware in a system can disrupt operations and the availability of information in networks while also jeopardizing the integrity and confidentiality of such information, which poses a grave issue for sensitive and critical operations. Traditional approaches to malware detection often used by antivirus software are not robust in detecting previously unseen malware. As a result, they can often be circumvented by finding and exploiting vulnerabilities of the detection system. This study involves using natural language processing techniques, considering the recent advancements made in the field in recent years, to analyze the strings present in the executable files of malware. Specifically, we propose a topic modeling-based approach whereby the strings of a malware’s executable file are treated as a language abstraction to extract relevant topics, which can then be used to improve a classifier’s detection performance. Finally, through experiments using a publicly available dataset, the proposed approach is demonstrated to be superior in performance to traditional techniques in its detection ability, specifically in terms of performance measures such as precision and accuracy.

Supervised machine learning depends on training a model to mimic previous labeled results. The problem with a small dataset is that data augmentation is necessary to increase the generalization of the model to future images, but we have observed that future images won’t necessarily be in the same domain as the augmented images. To alleviate this problem we applied image segmentation multiple times on the same image by using the same data augmentation techniques on the image in question, and then we merged the results using a priority based on the class weights used when training the model. Merging the segmentation results from the augmented images increased the mean-intersection-over-union over the inference results that used a single image.

Highly distributed connected systems, such as the Internet of Things (IoT), have made their way across numerous fields of application. IoT systems present a method for the connection for various heterogeneous devices across the internet, facilitating the efficient distribution, collection and processing of system-related data. However, while system inter connectivity has aided communication and augmented the effectiveness of integrated technology, it has also increased system vulnerability. To this end, researchers have proposed various security protocols and frameworks for IoT ecosystems. Yet while many suggested approaches augment system security, centralization remains an area of concern within IoT systems. Therefore, we propose the use of a decentralization scheme for IoT ecosystems based on Blockchain technology. The proposed method is inspired by Helium, a public wireless long-range network powered by blockchain. Each network node is characterized by its device properties, which are comprised of local and network-level features. Communication in the network requires the testimony of other companion nodes, ensuring that anomalous behaviour is not accepted and thereby preventing malicious attacks of various sorts.

Aviation fundamentally depends on well structured and rigorous record keeping frameworks. The demand for accurate data is continuously increasing for both operational and security processes, particularly in civil aviation. The principal challenge for civil aviation is protecting and sharing sensitive information, while guarding itself against the identity of illegitimate users. The emergence of blockchain and biometrics has given rise to a new direction of technological innovation, which has potential to solve the challenges fundamentally affecting the aviation sector. The proposed algorithm is designed to extract multimodal physiological biometrics, fusion with cryptographic hash to create 1024-bit (256 hex) hash in blockchain vector subspace. The primary challenge of this research is to incorporate biometrics attributes within blockchain hash function, making data available to all authorized aviation entities, and protecting sensitive PII Information against privacy, security, and unlinkability attacks. Blockchain and biometric systems provide security measures on their own. In conjunction with asymmetric encryption, the data becomes more secure when live human physiological attributes are required to unlock a secret key, as it will be incorporated with every transaction of civil aviation. The stochastic model will provide variabilities in the estimation of unique cipher key. Biometrics is the only part of the process of encrypting data which ensures unique digital identity with the presence of live users. On the other side, the time and difficulty of guessing a secret key along with global accessibility of universal data set in real time, are what makes the proposed model the most robust and cryptographically secure, integrated cybersecurity platform.

Blockchain applications go far beyond cryptocurrency. As an essential blockchain tool, smart contracts are executable programs that establish an agreement between two parties. Millions of dollars of transactions attract hackers at a hastened pace, and cyber-attacks have caused large economic losses in the past. Due to this, the industry is seeking robust and effective methods to detect vulnerabilities in smart contracts to ultimately provide a remedy. The industry has been utilizing static analysis tools to reveal security gaps, which requires an understanding and insight over all possible execution paths to identify known contract vulnerabilities. Yet, the computational complexity increases as the path gets deeper. Recently, researchers have been proposing ML-driven intelligent techniques aiming to improve the efficiency and detection rate. Such solutions can provide quicker and more robust detection options than the traditionally used static analysis tools. As of this publication date, there is currently no published survey paper on smart contract vulnerability detection mechanisms using ML models. In order to set the ground for further development of ML-driven solutions, in this survey paper, we extensively reviewed and summarized a wide variety of ML-driven intelligent detection mechanism from the following databases: Google Scholar, Engineering Village, Springer, Web of Science, Academic Search Premier, and Scholars Portal Journal. In conclusion, we provided our insights on common traits, limitations and advancement of ML-driven solutions proposed for this field.

The future of military combat brings a comprehensive suite of interconnected physical objects that are embedded with sensors, processing capabilities, and software that will exchange data in real time over both public facing internet services and dedicated military communications networks. New interconnected devices could include anything from combat gear embedded with biometric wearables, sensors for collection of imagery, audio, video, electromagnetic signals, chemical and biological agents, smart guns, and other military equipment that will advance that state of interconnected military arsenal. Adding to this evolving complexity will be new processing accelerators, distributed cloud environments, next generation cellular towers, distributed applications, sensing devices, and crowed-sourced intelligence we hope to leverage from the commercial sector. Predictive battlefield analytics and robust security strategies must be implemented for this ecosystem to be successful and not present a “weak link” for our adversary to exploit. Decentralization, low power consumption, and security are also vital to an Internet of Things (IoT) network architecture operating on the battlefield. In this investigative study, a hierarchical approach is explored. Principles of Zero-Trust which assume there is no implicit trust granted to assets based solely on their physical or network location are explored to ensure the robustness and security of the ecosystem. We explore a theoretical IoT network design using LoRaWAN (Long Range Wide Area Networks) and Distributed Ledger Technology (DLT) that is secure and decentralized while meeting low power requirements. To achieve this in a military operational environment, a private blockchain will be explored that has been developed by the Linux Foundation called the Hyperledger Fabric. This private blockchain based ecosystem provides a modular enterprise distributed framework with plug-and-play capabilities, decentralization, scalability, immutability, and a tailorable consensus mechanism. Integrating these capabilities will enable a more secure ecosystem of bi-directional communication, end-to-end security, and mobility.[1]