II FY2022 Events & Activities

Meet & Greet | June 1, 2022

GOAL: The Meet & Greets are used to introduce the group to AFRL and the II Staff. This session will be given on-site only. Presenters included: Dr Mark Linderman, RI Chief Scientist; Dr Stan Wenndt, II Lead; Dr John Lugansland, AFOSR; and Ms Heather Hage, President & CEO of the Griffiss Institute. Click on the link below to view and/or download the presentation.

AFRL/RI Overview

Welcome to the II

AFOSR Overview

Introduction to Innovare and GI


5-Minute Introductions | June 7, 2022

GOAL: The objective of the 5-minute introductions is to introduce the faculty members that are currently working at Rome. The goal of these quick overviews is to share each other’s research interests for potential collaborations. Click on the link below to view and/or download the presentation.

SFFP Participants

STFP Participants

VFRP Participants


Poster Board Session | July 27, 2022

GOAL: The objective of the Poster Board Session is to elicit interaction between the AFRL engineers and faculty members. The workshop was held on-site. The posters are organized by Core Technical Competency (CTC). Click on the link below to view and/or download the poster boards.

Autonomy, Command and Control and Decision Support (AC2)

Connectivity and Dissemination (CAD)

Cyber Science and Technology

Processing and Exploitation (PEX)


II Tech Talks

GOAL: The objective of the II Tech Talks is to introduce participants to various technical areas being worked on. These are to be technical presentations and last no longer the 30 minutes with questions. These presentations will be held both on-site and via Zoom.

Session 1 - June 16, 2022

Dr. Taeho JungDr. Taeho Jung
tjung@nd.edu
Download Slides

Secure Computation & Management for Large-scale Data Collection

Abstract: The world is increasingly digitized and “datafied” nowadays due to the proliferation of Internet-of-Thing (IoT) devices, and the collection of large-scale data has become pervasive. Though being useful and valuable, such a pervasive data collection has also brought a significant concern of security and privacy due to the growing data breaches, therefore there is an imminent need for techniques to secure the computation and management of the collected data. This talk will describe two threads of relevant research. First, I will introduce the integration of hardware-based secure computation and crypto-based secure computation that results in better efficiency than either one used alone. Second, I will describe how blockchain can be innovated to securely and efficiently store and manage the data records with dependencies while preserving log/linkage confidentiality.

Dr. Jie WeiDr. Jie Wei
jwei@ccny.cuny.edu
Download Slides

Multi-Modal Computing in Civil, Military & Medical Applications

Abstract: In this presentation, I will introduce my recent work on using state-of-the-art AI/ML approaches in civil engineering, military applications, and medical computing using a wide variety of data modalities. In civil engineering, buildings, bridges, and highways can be inspected and classified using vibration-based sensors and AI methods for remote inspection and monitoring; in military applications, in collaboration with AFRL, AFOSR, and ARO, difficult object classification and threat detection using multiple sensors were achieved with encouraging performance; in medical imaging and cancer treatment planning/delivery, by joining forces with biomedical engineers and medical physicians in hospitals and cancer centers, lung/liver cancer treatment planning and treatment delivery based on surface geometry and motion features developed by cutting-edge ML approaches yield valuable non-invasive approaches, recent work on sleep staging and event classifications based on ensemble learning achieved human expert-level performance.

 

Session 2 - June 23, 2022

Dr Kevin Kwiat

Dr. Kevin Kwiat
kwaitk@gmail.com
Download Slides

A Technique for Fighting-Through Data Leakage Hardware Trojans: From Concept to Physical Instantiation

Abstract: Many integrated circuit (IC) design houses continue to outsource the production of their chips to other countries. This creates an on-going opening for cyber-attacks: when a design is sent out to be manufactured, the trustworthiness of the resultant IC can no longer be guaranteed because it is possible to insert hardware Trojans directly into the IC during its design and manufacturing process. If the insider threat is considered, then on-shore production can also be the target for hardware Trojans. For many in the cyber assurance arena confidentiality is paramount, so the possibility of a hardware Trojan leaking sensitive data is of greatest concern. To undermine in-situ data-leakage hardware Trojans, the U.S. Air Force has patented and patent-pending inventions covering 1) combinational and 2) sequential logic. This presentation will focus on the combinational method (the sequential version is similar). The inventions, having now been licensed, defeat data leakage through a randomized encoding and split manufacturing technique. Since its conception the technique has been elevated from the IC level to the printed circuit board level and is demonstrable in hardware.

Dr. Jeremy Chapman

Dr. Jeremy Chapman
jeremy.chapman.6@us.af.mil
Download Slides

Conceptual Spaces

Abstract: Conceptual Spaces are a new and emerging form of cognitive modeling that represents an alternative to tradition neural network machine learning techniques. They work by building a geometric space for representing how the human mind perceives concepts. They are capable of higher-level data fusion and fusion of both hard (e.g. measurable) and soft (e.g. contextual) data. An explanation of Conceptual Spaces as well as their underlying mathematics will be address in this technical briefing as well as a use case involving the threat of a space-to-space physical attack, known as a spacecraft kinetic kill, with the overall goal of improving Space Situational Awareness (SSA).

 

Session 3 - June 30, 2022

Dr. Houbing SongDr. Houbing Song
SONGH4@erau.edu
Download Slides

AI for Cybersecurity and Security of AI

Abstract: The mutual needs and benefits of AI and cybersecurity have been widely recognized. AI techniques are expected to enhance cybersecurity by assisting human system managers with automated monitoring, analysis, and responses to adversarial attacks. Conversely, it is essential to guard AI technologies from unintended uses and hostile exploitation by leveraging cybersecurity practices. The interplay between AI/machine learning, and cybersecurity introduces new opportunities and challenges in the security of AI as well as AI for cybersecurity. In this talk, I will present my recent research on AI for cybersecurity and the security of AI. First, I will introduce my research on AI for cybersecurity, i.e., real-time machine learning for quickest event (threat/intrusion/vulnerability…) detection. Next, I will present my research on the security of AI, i.e., coverage-driven testing and evaluation of deep learning systems.

Dr. John LicatoDr. John Licato
licato@usf.edu
Download Slides

Why is it so Hard for AI to Reason About Regulations?

Abstract: It is easy to assume that artificially intelligent systems because they are programmed in languages that follow rigid rules, can easily follow any rules we give them. But as it turns out, they have immense difficulty properly interpreting many of the rules, laws, regulations, guidelines, and codes of behaviors we humans use daily. This is because those things make heavy use of open-textured terms, and even our best AI has trouble interpreting such terms in human-like ways. I describe this problem, implications for DoD research, and progress made in addressing this.

 

Session 4 - July 7, 2022

Dr. Fanxin Kong

Dr. Fanxin Kong
fkong03@syr.edu
Download Slides

Attack-Resilient Cyber-Physical Systems

Abstract: Compared to conventional IT systems, challenges in CPS security are distinct in terms of not only consequences in case of security breaches but also attack surfaces. Sensors act as an interface between the cyber and physical space, and thus their data integrity is critical to CPS security. In this talk, Dr. Fanxin Kong will present his recent works on real-time sensor attack detection and real-time recovery from attacks. These works are motivated by the fact that the timing correctness of attack defense receives considerably less attention than its functional correctness, but untimely defense is just damaging. The proposed solutions cover formal methods and machine learning techniques.

Dr. Jiafeng (Harvest) Xie

Dr. Jiafeng Harvest Xie
Jiafeng.xie@villanova.edu
Download Slides

Multi-Modal Computing in Civil, Military & Medical Applications

Abstract: In this presentation, I will introduce my recent work on using state-of-the-art AI/ML approaches in civil engineering, military applications, and medical computing using a wide variety of data modalities. In civil engineering, buildings, bridges, and highways can be inspected and classified using vibration-based sensors and AI methods for remote inspection and monitoring; in military applications, in collaboration with AFRL, AFOSR, and ARO, difficult object classification and threat detection using multiple sensors were achieved with encouraging performance; in medical imaging and cancer treatment planning/delivery, by joining forces with biomedical engineers and medical physicians in hospitals and cancer centers, lung/liver cancer treatment planning and treatment delivery based on surface geometry and motion features developed by cutting-edge ML approaches yield valuable non-invasive approaches, recent work on sleep staging and event classifications based on ensemble learning achieved human expert-level performance.

 

Session 5 - July 14 2022

Dr. Javad Mohammadi

Dr. Javad Mohammadi
javadm@utexas.edu
Download Slides

Multi-agent Inference and Decision Making in Networked Cyber-Physical Systems

Abstract: The digital transformation of cyber-physical infrastructures is driven by connectivity and autonomy. The consequence of this transformation is that the system’s physical structure has become significantly more distributed, the number of cyber-enabled autonomous components has dramatically increased, and the inter-infrastructure interdependence has expanded. These cyber and physical transitions make system-wide inferences and decision makings challenging.

The energy system is an example of an infrastructure that is increasingly facing these challenges. The future electric infrastructure will differ from the current system by the increased integration of smart grid-interactive buildings, decentralized energy generation, widespread energy storage, communications, and sensing technologies. These advancements, combined with climate change concerns, and resiliency needs, are resulting in a more distributed and interconnected electric infrastructure, requiring decisions to be made at scale. The current centralized sensing and decision-making structures are not suitable to operate such highly decentralized systems.

In this talk, I will consider multi-agent information processing problems and introduce a set of privacy-preserving, adaptive, and computationally efficient solutions. I will showcase applications in addressing the decarbonization and resiliency needs of energy systems. The proposed solutions are based on the consensus + innovations multi-agent framework, in which the consensus term enforces agreement among agents while the innovations updates ensure that agents’ constraints are satisfied.

Dr. Yu Chen

Dr. Yu Chen
ychen@binghamton.edu
Download Slides

Privacy Preserving Decentralized Smart Public Safety Surveillance

Abstract: The recent advancements in the Internet of Things (IoT) and Edge-Fog-Cloud Computing technologies make the Smart Public Safety Surveillance (SPSS) system feasible for Smart Cities. However, existing monolithic service-oriented architecture (SOA) is unable to provide scalable and extensible services in large-scale heterogeneous networks. In addition, traditional centralized security solutions not only can be a performance bottleneck or suffer the single point of failure, but it also incurs privacy concerns on improperly use of private information. In this talk, a privacy-preserving, secure, and decentralized SPSS system is introduced. Inspired by blockchain and microservices technologies, the BLockchain-Enabled Decentralized Smart Public Safety (BlendSPS) system tackles three challenging issues at the edge: (1) fake/false frame injection attacks, (2) data integrity in transmission, and (3) privacy violation in surveillance. A microservices based hybrid blockchain fabric enables decentralized security architecture, which supports immutability, auditability, and traceability for secure data sharing and operations among participants. An extensive experimental study verified the feasibility of BlendSPS in edge networks.

 

Session 6 - July 21 2022

Dr. Ji Liu

Dr. Ji Liu
ji.liu@stonybrook.edu
Download Slides

Decentralized Multi-Armed Bandits

Abstract: Multi-armed bandit (MAB) is a fundamental reinforcement learning problem which exemplifies the exploration-exploitation trade-off as a sequential decision-making process. This talk will focus on decentralized MAB problems over a network of multiple agents, each of which can communicate only with its neighbors, where neighbor relationships are described by a possibly time-varying graph. Each agent makes a sequence of decisions on selecting an arm from a given set of candidates, yet it only has access to local samples of the reward for each action, which is a random variable. Two cases will be discussed. First, when the agents have heterogeneous observations of rewards, their goal is to minimize the cumulative expected regret with respect to the true rewards of the arms, where the mean of each arm’s true reward equals the average of the means of all agents’ observed rewards. It is shown that for any uniformly strongly connected graph sequence, a decentralized MAB algorithm achieves guaranteed regret for each agent at the optimal order. Second, when all the agents share a homogeneous distribution of each arm reward, two decentralized MAB algorithms are proposed, respectively based on the classic upper confidence bound (UCB) algorithm and the state-of-the-art KL-UCB algorithm. It is shown that they both guarantee each agent to achieve a better logarithmic asymptotic regret than their single-agent counterparts, provided that the agent has at least one neighbor. The algorithms can be further tailored to be fully resilient to adversaries and malicious attacks capable of introducing untrustworthy information into the communication network, built upon neighbor redundancy. All proposed algorithms are fully decentralized without using any network-wide information.

Dr. Sanjay Madria

Dr. Sanjay K. Madria
madrias@mst.edu
Download Slides

Data Collection in IoT Networks using Trajectory Encoded with Geometric Shapes for IoT Sensing Applications

Abstract: The Mobile Edge Computing (MEC) mitigates the bandwidth limitation between the edge server and the remote cloud by directly processing the large amount of data locally generated by the network of the internet of things (IoT) at the edge. To reduce redundant data transmission, In this work, we proposed a data collection scheme that only gathers the necessary data from IoT devices along a trajectory. Instead of using and transmitting location information (to preserve location anonymity), a virtual coordinate system called "distance vector of hops to anchors" (DV-Hop) is used. The proposed trajectory encoding algorithm uses ellipse and hyperbola constraints to encode the position of interest (POI) and the trajectory route to the POI. Sensors make routing decisions only based on the geometric constraints and the DV-Hop information, both of which are stored in their memory. The proposed DV-Hop updating algorithm enables the users to collect data in an IoT network with mobile nodes. The experiments show that in heterogeneous IoT networks, the proposed data collection scheme outperforms two other state-of-the-art topology-based routing protocols, called ring routing, and nested ring. The results also show that the proposed scheme has better latency, reliability, coverage, energy usage, and provide location privacy compared to state-of-the-art-schemes. The proposed scheme can be used in many IoT and Sensing application including non-GPS environment.

 

Session 7 - July 28, 2022

Dr. Helen Durand

Dr. Helen Durand
helen.durand@wayne.edu
Download Slides

Next-Generation Manufacturing for Nonlinear Processes: Profits, Cybersecurity, and Quantum Computing

Abstract: Next-generation manufacturing systems will have greater autonomy and efficiency due to advances in computers, control designs, and networking that enable more data to be utilized from throughout a plant and promote more optimal decision-making.  While these advances provide these gains for process operation, they also introduce more routes by which an attacker could compromise an industrial control system, affecting company profits and potentially system safety.  Furthermore, despite the significant advances in computing power over the last decades, many complex engineering problems remain time-consuming to solve on classical computing devices, such as computational fluid dynamics and finite element analysis models.  This limits the complexity of models that can be considered when designing and evaluating control strategies and raises the question of whether quantum computers could hold any benefits for reducing computation time for control-relevant problems in the future.  The means for addressing this latter question is, however, non-obvious, so that initially exploring how control laws can be implemented on quantum computers may aid in providing direction toward answering the broader question.  In this talk, we will discuss our work in designing cyberattack detection policies for industrial control systems, and our preliminary work in evaluating impacts of the imperfections in today's quantum computers on the success of implementing control laws on these devices.  We will discuss control-theoretic guarantees that can be made when cyberattacks occur but are not detected through the co-design of the detection strategies and the control systems and will also discuss preliminary results for quantum computing-implemented control that suggest that there may be conditions under which calculating control actions with the aid of such computers can be stabilizing despite the imperfections in the devices.

Dr. Hisham Kholidy

Dr. Hisham Kholidy
madrias@mst.edu
Download Slides

Toward Zero Trust Security in 5G Networks: A Smart Contract Approach

Abstract: A 5G network would dramatically improve military communication and situational awareness. Developing trust in 5G network slicing is an important issue since as mobile networks evolve the number of internal components, network functions, and the use of virtualized elements increase, and more actors and stakeholders involve in multiple interactions. The current security practices that are based on a perimeter/defense-in-depth cybersecurity approach are proven to be ineffective to address the current and future cybersecurity challenges because they assume that everything on the inside of a network is trustworthy. One promising approach of growing significance in the telecom security sphere is the Zero-Trust (ZT) security model. The ZT is the term for an evolving set of cybersecurity paradigms that move network defenses from static, network-based perimeters to focus on users, assets, and resources. In this presentation, we will highlight our proposed approach that enables trustworthy deployment and management of network slices in a 5G core network. The proposed approach incorporates trust in brokering architecture allowing the slice provider to securely create end-to-end network slices while outsourcing resources from different infrastructure providers. A 5G network would dramatically improve military communication and situational awareness. Developing trust in 5G network slicing is an important issue since as mobile networks evolve the number of internal components, network functions, and the use of virtualized elements increase, and more actors and stakeholders involve in multiple interactions. The current security practices that are based on a perimeter/defense-in-depth cybersecurity approach are proven to be ineffective to address the current and future cybersecurity challenges because they assume that everything on the inside of a network is trustworthy. One promising approach of growing significance in the telecom security sphere is the Zero-Trust (ZT) security model. The ZT is the term for an evolving set of cybersecurity paradigms that move network defenses from static, network-based perimeters to focus on users, assets, and resources. In this presentation, we will highlight our proposed approach that enables trustworthy deployment and management of network slices in a 5G core network. The proposed approach incorporates trust in brokering architecture allowing the slice provider to securely create end-to-end network slices while outsourcing resources from different infrastructure providers.

 

Session 8 - August 4, 2022

Dr. Dakai Zhu

Dr. Dakai Zhu
Dakai.Zhu@utsa.edu
Download Slides

Low-Power Dependable Computing and Beyond

Abstract: Learning-enabled components (LEC) have been gradually incorporated in safety-critical systems, which introduces more computational demand with additional power consumption. However, it has been shown that the widely deployed power management technique, Dynamic Voltage and Frequency Scaling (DVFS), has direct and negative effects on system reliability, where the problem becomes more prominent as the technology size scales down and more transistors are integrated on a chip. This poses a serious challenge on system reliability and demands novel techniques that can address both energy consumption and reliability simultaneously. Here, I will first introduce the Reliability-Aware Power Management (RAPM) framework, which systematically exploits slack time to minimize energy consumption while preserving system reliability within various timing requirements, for both uniprocessor and multi-core/processor systems. Then, considering heterogeneous computing platforms, I will present our recent research results on exploiting their features to tackle energy constrained reliability issues. Finally, I will share some ideas on supporting robustness of learning components in safety-critical systems running on heterogeneous computing platforms.

Dr. Qiang Ji

Dr. Qiang Ji
jiq@rpi.edu
Download Slides

Knowledge Augmented Deep Visual Learning

Abstract: Substantial progresses have been made in computer vision recently because of the latest algorithmic advances in deep learning.  Despite these successes, current computer vision algorithms are data-driven, require a large amount of annotated data to perform well, and do not generalize well to novel data/tasks.  To address these issues, we propose to augment current deep visual learning algorithms with well-established generic prior knowledge to achieve data-efficient and generalizable visual learning.

 Specifically, we propose to investigate three technical issues: knowledge identification, knowledge representation, and knowledge encoding.  For knowledge identification, we systemically identify prior knowledge from well-established domain theories or extensive studies that govern the properties of the target variables or the underlying data generation mechanism.  Knowledge representation will focus on developing different representations to accurately capture the identified knowledge.  We will use mathematical equations, probabilistic constraints, and pseudo-data to represent knowledge of different types.   For knowledge encoding, we introduce different encoding schemes, including customized architectures, model learning regularization, and data augmentation, to systematically incorporate knowledge into different stages of visual learning to achieve rigorous integration of knowledge with data.  Through the proposed framework, the prior knowledge and data are effectively integrated and they work synergically to gain vision algorithms that are data efficient, robust, and generalizable across datasets/domains.   To demonstrate the proposed framework, we apply it to different computer vision tasks for human behavior analysis and recognition, including facial expression recognition, 3D body mesh reconstruction, and human action/gait recognition.

 

Session 9 - August 11, 2022

Dr. Chin-Tser Huang

Dr. Chin-Tser Huang
HUANGCT@cse.sc.edu
Download Slides

Privacy-Preserving Consensus Based on Secure Multiparty Computation for IoT Environment

Abstract: With the rapid advancement and wide application of blockchain technology, blockchain consensus protocols, which are the core part of blockchain systems, along with the privacy issues, have drawn much attention from researchers. A key aspect of privacy in the blockchain is the sensitive content of transactions in the permissionless blockchain. Meanwhile, some blockchain applications, such as cryptocurrencies, are based on low-efficiency and high-cost consensus protocols, which may not be practical and feasible for other blockchain applications. In this talk, we introduce an efficient and privacy-preserving consensus protocol, called Delegated Proof of Secret Sharing (DPoSS), which is inspired by secure multiparty computation. Specifically, DPoSS first uses polynomial interpolation to select a dealer group from many nodes to maintain the consensus of the blockchain system, in which the dealers in the dealer group take turns to pack the new block. In addition, since the content of transactions is sensitive, our proposed design utilizes verifiable secret sharing to protect the privacy of transmission and defend against the malicious attacks. Extensive experiments show that the proposed consensus protocol achieves fairness during the process of reaching consensus.

Dr. Hong Zhao

Dr. Hong Zhao
zhao@fdu.edu

A Hardware Security Approach for FPGA-based Embedded System

Abstract: Security has been a concern for all connected devices. Attackers continually search for vulnerabilities from software, firmware, and all the way down to hardware level. At the same time, cyber security has also been pushed to the hardware platform to keep invaders out. Hardware enabled security solutions can provide a stronger foundation than one offered by software, or firmware which can be modified with relative ease. With FPGA increasingly being used as a hardware platform for more critical applications, providing FPGA-based security approach becomes a necessity. The project addresses SRAM based FPGA security weak point and focuses on providing cost effective solution to its configuration bitstream confidentiality/ authentication, and cyber resilience at platform level with emphasis on preventing stage.  Applying hardware security primitive PUF for providing secrecy to the proposed security approach makes key management feasible for remotely placed devices.

 

Session 10 - August 18, 2022

Dr. Xiaokang Qiu

Dr. Xiaokang Qiu
xkqiu@purdue.edu
Download Slides

Toward More Accessible and Trustworthy Program Synthesis

Abstract: Program synthesis is the process of automatically generating programs that meet the user’s intent. In the last two decades, this traditional research area has experienced a renaissance and witnessed emerging programming paradigms, automated tools, and prominent applications. However, the promising progress also exposes a tension between making program synthesis more user-friendly and guaranteeing the quality of the produced code.

 

In this talk, I will discuss two projects from our work on pushing further toward more accessible and trustworthy program synthesis. First, I will present Comparative Synthesis, an interactive synthesis framework that learns near-optimal programs through comparative queries, without explicitly specified optimization targets. We develop a voting-guided learning algorithm which provides a provable guarantee on the quality of the found program. We have implemented this approach in a system Net10Q for wide-area network design. Experiments with oracles and a pilot user study with network practitioners and researchers show Net10Q is effective in finding allocations that meet diverse user policy goals in an interactive fashion. Second, I will present Cooperative Synthesis, a framework for solving Syntax-Guided Synthesis (SyGuS) problems, i.e., finding a program satisfying semantic specification as well as user-provided grammar. The framework repeatedly splits large SyGuS problems into subproblems and solves them by deduction or enumeration separately. Cooperative synthesis has been embodied in DryadSynth, a SyGuS solver which won the CLIA track of the SyGuS competition two years in a row.

Dr. James Xiaojiang Du

Dr. James Xiaojiang Du
dxj2005@gmail.com
Download Slides

Anomaly Detection and Prevention for Smart Internet of Things (IoT) Systems

Abstract: As IoT devices are integrated via automation and coupled with the physical environment, anomalies in smart environments (e.g., smart homes, smart buildings, smart bases) whether due to attacks, device malfunctions or human mistake, may lead to severe consequences. Prior works that utilize data mining techniques to detect anomalies suffer from high false alarm rates and missing many real anomalies. Our observation is that data mining-based approaches miss a large chunk of information about automation programs (also called smart apps) and devices. We design Home Automation Watcher (HAWatcher), an anomaly detection and prevention system for smart environments. This work has been published by one of the top four security conferences - USENIX Security 2021 (the acceptance rate was 17%). In this talk, I will present HAWatcher and some other work from my group on IoT security and privacy.