Computer Sciences Commons^™

Understanding The Oss Communities Of Deep Learning Frameworks: A Comparative Case Study Of Pytorch And Tensorflow, Yunqi Chen, Zhiyuan Wan, Yifei Zhuang, Ning Liu, David Lo, Xiaohu Yang

Research Collection School Of Computing and Information Systems

Over the past two decades, deep learning has received tremendous success in developing software systems across various domains. Deep learning frameworks have been proposed to facilitate the development of such software systems, among which, PyTorch and TensorFlow stand out as notable examples. Considerable attention focuses on exploring software engineering practices and addressing diverse technical aspects in developing and deploying deep learning frameworks and software systems. Despite these efforts, little is known about the open source software communities involved in the development of deep learning frameworks. In this article, we perform a comparative investigation into the open source software communities of …

Varium: Variational Autoencoder For Multi-Interest Representation With Inter-User Memory, Nhu Thuat Tran, Hady W. Lauw

Research Collection School Of Computing and Information Systems

Frameworks for discovering multiple user interest factors based on Variational AutoEncoder (VAE) has demonstrated competitive recommendation performance. However, as VAE only considers one user as input at a time, sharing across like-minded users may not be adequately facilitated. Moreover, interest sharing between users is not always available and thus, poses a challenge for VAE to explicitly model this information. To resolve this, we introduce an inter-user memory-based mechanism to unsupervisedly discover latent interest sharing between users under VAE framework. Concretely, we design a memory including an array of prototypes, each hypothetically representing a group of users sharing a particular interest. …

Multi-Uav Reconnaissance Mission Planning Via Deep Reinforcement Learning With Simulated Annealing, Mingfeng Fan, Huan Liu, Guohua Wu, Aldy Gunawan, Guillaume Sartoretti

Research Collection School Of Computing and Information Systems

Unmanned aerial vehicles (UAVs) are widely used in reconnaissance missions due to their autonomy and flexibility. Efficient mission planning for multiple UAVs is crucial for tasks such as traffic monitoring and data collection. However, existing approaches to multi-UAV reconnaissance mission planning problem (MURMPP) often struggle with high computational demands, leading to suboptimal solutions. To overcome this challenge, we introduce a divide-and-conquer framework that splits the problem into two phases: target allocation and UAV routing, effectively reducing computational complexity. Specifically, we propose a hybrid method, SA-NNO-DRL, which combines the nearest neighbor optima-based deep reinforcement learning (NNO-DRL) approach with simulated annealing (SA). …

Retrieval Augmented Recipe Generation, Guoshan Liu, Hailong Yin, Bin Zhu, Jingjing Chen, Chong-Wah Ngo, Yu-Gang Jiang

Research Collection School Of Computing and Information Systems

The growing interest in generating recipes from food images has drawn substantial research attention in recent years. Existing works for recipe generation primarily utilize a two-stage training method—first predicting ingredients from a food image and then generating instructions from both the image and ingredients. Large Multi-modal Models (LMMs), which have achieved notable success across a variety of vision and language tasks, shed light on generating both ingredients and instructions directly from images. Nevertheless, LMMs still face the common issue of hallu- cinations during recipe generation, leading to suboptimal performance. To tackle this issue, we propose a retrieval augmented large multimodal …

Gender Biases Within Artificial Intelligence And Chatgpt: Evidence, Sources Of Biases, And Solutions, Qi Hui Jerlyn Ho, Andree Hartanto, Andrew Koh, Nadyanna M. Majeed

Research Collection School of Social Sciences

The growing adoption of Artificial Intelligence (AI) in various sectors has introduced significant benefits, but also raised concerns over biases, particularly in relation to gender. Despite AI's potential to enhance sectors like healthcare, education, and business, it often mirrors reality and its societal prejudices and can manifest itself through unequal treatment in hiring decisions, academic recommendations, or healthcare diagnostics, systematically disadvantaging women. This paper explores how AI systems and chatbots, notably ChatGPT, can perpetuate gender biases due to inherent flaws in training data, algorithms, and user feedback loops. This problem stems from several sources, including biased training datasets, algorithmic design …

Abolition By Algorithm, Peter N. Salib

Michigan Law Review

In one sense, America’s newest abolitionist movement—advocating the elimination of policing and prison—has been a success. Following the 2020 Black Lives Matter protests, a small group of self-described radicals convinced a wide swath of ordinary liberals to accept a sweeping claim: Mere reforms cannot meaningfully reduce prison and policing’s serious harms. Only elimination can. On the other hand, abolitionists have failed to secure lasting policy change. The difficulty is crime. In 2021, following a nationwide uptick in homicides, liberal support for abolitionist proposals collapsed. Despite being newly “abolition curious,” left-leaning voters consistently rejected concrete abolitionist policies. Faced with the difficult …

Turbulence Prediction Using Non-Linear Phase Space Analysis, Jeremy Quijano

Shelby Hall Graduate Research Forum Posters

Our research presents a novel approach for turbulence prediction in computational fluid dynamics (CFD) simulations using a non-linear phase space analysis (NLPSA) and threshold algorithm. NLPSA has been utilized in medical applications to predict seizures, as well as in cybersecurity to detect malicious control and utilization of computing systems. NLPSA uses time-series data to learn the normal operating state of the system, then sets a threshold to predict when the system becomes abnormal. Turbulence prediction is similar, such that a fluid system changes from normal to abnormal. Turbulence prediction methods currently utilize machine learning tools, such as convolutional neural networks …

Out-Of-Band Anomaly Detection For Real Time Operating Systems, Jeff K. Holifield

Shelby Hall Graduate Research Forum Posters

Real Time Operating Systems (RTOS) are increasing present throughout the industrial, business, defense, and healthcare spaces. These lightweight and efficient operating systems are designed to run on embedded, resource constrained devices, often within cyber-physical systems (CFS). A defining characteristic of RTOSs is that they are deterministic. Tasks are scheduled to run on fixed timelines within guaranteed execution windows. To accomplish tasks on time, real time software must conform to worst case execution times (WCETs) as design parameters. WCET is the maximum time a particular task can take to complete. Exceeding the WCET could cause system failure and lead to damage, …

A Framework For Design Recovery, Eric Diep

Shelby Hall Graduate Research Forum Posters

Due to the increase in diverse chip production over the past decade, reverse engineering has become a difficult and daunting task. This research will create a guideline for methods to achieve design recovery of microchip logic. To accomplish this, we plan on using hardware tools such as laser delayering and microscopy imaging. We will be focusing on the DA14580 Dialog semiconductor, commonly implemented in tile trackers, to extract information for design recovery. This implementation technique is novel due to tools that have not been performed with this type of microchip. The purpose for this research is to create a framework …

Analysis Of Forensic Techniques For Additive Manufacturing Devices, Daniel B. Miller, Brad Glisson, Mark Yampolskiy, J Todd Mcdonald

Shelby Hall Graduate Research Forum Posters

Additive Manufacturing (AM) is a set of newer computer-dependent production technologies that is seeing rapid adoption across a wide variety of industries, including defense, aerospace, automotive, and healthcare. With increased adoption comes an increased opportunity for misuse and abuse of such systems, which will lead to an increased need for Digital Forensic investigations into these platforms. This research forensically analyzes a number of AM devices to explore the options available for data acquisition as well as the impacts of hardware and software design choices on the analysis and investigation results. Hardware is investigated using Open-Source Intelligence (OSINT) sources to determine …

Polyglot File Detection For Forensics, Chase Stevens, Michael Black

Shelby Hall Graduate Research Forum Posters

Polyglot files are problematic as payloads can be hidden inside them while simultaneously evading discovery by current forensic tools. Autopsy, one of the most used forensic tools in investigations, uses known signatures of file types (e.g., headers, trailers) to identify and recover files. Polyglot files are a unique case in which files are intentionally combined with other file types to make them appear benign while still containing malicious payloads. Polyglots are possible due to combinations being considered valid in both formats. Polyglot files inherently evade detection from signature-based forensic tools and malware scanners because the tools and malware scanners are …

Using Image-Based Representation For Network Intrusion Detection, Chakriya Suon, J. Todd Mcdonald

Shelby Hall Graduate Research Forum Posters

The primary focus of our research is to evaluate the effectiveness of converting network traffic data, PCAPs, into image-based representations for anomaly-based network intrusion detection. We aim to analyze PCAPs to detect malware, or malicious software in hopes of creating a useful approach for anomaly detection against cyber threats including Advanced Persistent Threats (APTs). With the rise of cyber threats, cybersecurity continues to play a critical role in the ever-changing landscape of technology, by protecting and defending against threat agents. Our research will apply novel machine learning (ML) techniques to detect potential malware transmitted over a network effectively. The overall …

Learning Without Labels: A Self-Supervised Learning Approach For Anomaly Detection In Control Systmes, Barbara Gladney

Shelby Hall Graduate Research Forum Posters

Oil pipelines, water plant systems, and other critical infrastructure are managed and operated by industrial control systems (ICS). These systems safeguard the operations of critical infrastructures, requiring minimal disruption from cyberattacks or malfunctions. The use of anomaly detection methods in control systems (ICS) can reduce system interruptions. However, anomaly detection methods often require annotated data, which may not be available for the control system. Additionally, the datasets used for the control systems do not include sensor outputs and environmental data, resulting in a restricted view of the system. This research investigates how SSL models can be applied to different control …

Directing Sophisticated Cyberattacks On Public Water Infrastructure, Ayrton C. Purdy, Jordan Shropshire

Shelby Hall Graduate Research Forum Posters

In recent years there has been an increasing number of cyberattacks on public water generation and distribution systems. Advanced persistent attackers could usurp sensor and control systems to contaminate public drinking water. In order to conceal their malicious activity, they can manipulate sensor data flows to give the appearance of normal activity. The compromised sensors would report normal chemical levels even though unsafe water is entering the distribution system. In response, this research proposes a multi-sensor, cross-validation approach to anomaly detection. The proposed approach is designed to detect sophisticated cyberattacks which are not easily detectable using traditional cyber tools. The …

Security Vulnerabilities Of A Field Programmable Gate Array, Kylie Arnett

Shelby Hall Graduate Research Forum Posters

The primary goal of this research is to understand and exploit the security vulnerabilities of a Field Programmable Gate Array (FPGA), at the bitstream level. This paper is working to successfully show that an FPGA can be altered via the bitstream file, and a Trojan can be inserted into the device. Once a Trojan is successfully inserted into the FPGA and activated through a certain input value, a Siamese Neural Network (SNN) will be used to test the effectiveness of Trojan detection. Followed by recording the successful flag rate to detect Trojans, which will be averaged to determine the accuracy …

Establishing A Framework For Evaluating Machine Learning Performance And Security Across Computational Ecosystems, Krista Stacey, Todd R. Andel

Shelby Hall Graduate Research Forum Posters

The rapid evolution of computational ecosystems—ranging from embedded systems and cloud platforms to hybrid and quantum architectures—has introduced new challenges in deploying machine learning (ML) applications. While cloud computing offers scalability, it comes with increased latency and security risks, whereas edge computing, such as FPGA-based systems, provides real-time processing with constrained resources. Hybrid and quantum ecosystems further complicate decision-making, requiring careful trade-offs between performance and security. This research seeks to establish a framework for evaluating ML performance and security risks across these ecosystems, forming the foundation of the Computational Performance And Security System (COMPASS) decision-support tool. The study will systematically …

Application Of Graph Neural Networks With Phase Space Graphs, Parker H. Cole, Ryan Benton, Ralf Riedel, David Bourrie

Shelby Hall Graduate Research Forum Posters

Non-linear phase-space analysis models data represented as a graph transitioning between states in the time domain. By studying data transitions, we can predict the time a particular behavior occurs and classify the events (states) in a system. For example, we could classify neurological sensor data to determine if a person is asleep (state), or predict the direction in which a stock will move (transitions) based on micro trade patterns.

Previous research has demonstrated success in phase-space graphs in classifying malware, detecting network intrusions, and predicting seizures. However, the solutions either require calculating global graph features as inputs to a classifier, …

Development Of An Algorithm To Identify The Presence Of Luks-Encrypted Volumes On A Forensic Image Of A Drive, Nicholas Flynn, Michael Black

Shelby Hall Graduate Research Forum Posters

Current forensic tools struggle to effectively detect encrypted storage media. In recent years, there have been significant advancements, but a noticeable gap remains when it comes to identifying encrypted volumes from metadata alone. The goal of this research is to develop a novel algorithm that will identify the presence of volumes on a disk image which have been encrypted with the Linux Unified Key Setup (LUKS) encryption algorithm, in an effort to aid digital forensics investigations. Bad actors often use encryption as an anti-forensics tool to pose significant challenges to forensic investigators, especially when it is done within sections of …

Preserving Privacy In Senior Care At Home Monitoring Systems, Sam Russel, Ryan Benton, Amy Campbell, Scott Sittig

Shelby Hall Graduate Research Forum Posters

Many seniors prefer to live at home which necessitates research into the application of technologies to provide a safer environment with less caregiver resources. However, the application of home health care (HHC) monitoring for seniors is still in an evolutionary stage. Present HHC systems are produced by private companies with general regulatory guidelines lacking specific care of the elderly. As such, each company that produces such a system claims to have better safety, privacy, and security that their competitors. A pressing issues is devising and applying a general framework for the application of technologies that delivers safety while preserving privacy. …

False Narratives, Real Consequences, Russell W. Cantrell, Matt Campbell

Shelby Hall Graduate Research Forum Posters

Social media is an increasingly significant tool in modern cyber warfare, capable of rapidly shaping public opinion. The swift dissemination of information complicates efforts to distinguish fact from fiction [1]. During public health crises, healthcare professionals use these platforms to share updates, yet their credible content must contend with false or deliberately misleading narratives [2]. This environment creates an opportunity for cyberattacks through social media influence campaigns [3]. While disinformation's role in political interference has been widely studied, its potential to destabilize healthcare remains largely unexplored. Prior research primarily focuses on how vaccine misinformation affects the general public [4]. This …

Detecting Sensor Data Manipulation, Ricky Green, Michael Black

Shelby Hall Graduate Research Forum Posters

The integration of Information Technology (IT) and Operational Technology (OT) have made OT devices vulnerable to threats that have been successfully exploited with devastating results. Many modern techniques for hardening and securing enterprise IT systems are either incompatible with OT components in an Industrial Control System (ICS), reduce the efficiency of processes, or are prohibitively expensive to implement. Research in the area of ICS security focuses on a top-down approach, such as intrusion prevention by securing the perimeter of the network and hardening computer systems. This approach is useful in business IT systems, but full compatibility with OT components in …

Topical Text Segmentation For Stream Of Consciousness Writing, Yuwei Lu, Ryan Benton

Shelby Hall Graduate Research Forum Posters

Stream of consciousness writing has a long history, including novelists James Joyce and Virginia Woolf. However, there has been little work done in automated and semi-automated analysis of such writing, which is the focus of this work. We plan to divide real streams of consciousness writing into distinct topical units and then capture different momentary meaningful topics from these units. By doing this, researchers and readers could gain a more nuanced understanding of the narrative structure and thematic elements. In addition, it would also support applications in fields like psychology and linguistics, where understanding thought processes and narrative structures is …

Modelled Flooding Impacts On Lower Fish River Watershed, Sebastian Loschner

Shelby Hall Graduate Research Forum Posters

This study investigates the impacts of compound flooding in the Lower Fish River watershed, Baldwin County, Alabama, with a focus on the potential effects of sea level rise due to climate change. Coastal flooding, particularly in smaller watersheds, is a growing concern as it results from the interaction of multiple factors, including rainfall, tidal changes, and extreme weather events. Compound flooding, which involves multiple flood drivers, is expected to worsen with climate change, as increased precipitation and rising sea levels create heightened flood risks. However, existing research on compound flooding predominantly focuses on large-scale watersheds, leaving a knowledge gap in …

Using Machine Learning Models To Improve The Cyber Physical Security Of Drones, Sean Lee, Aviv Segev

Shelby Hall Graduate Research Forum Posters

This research proposes a new manner of implementing machine learning models such that, when applied on a drone, it will be able to accurately identify and maintain the authenticity of the entity sending the control data to the drone. To begin with, the drone will, for a pre-determined amount of signals received per unit time, determine the average signal strength (RSSI) of them and use that average to determine the approximate distance between the drone and the source of those signals. This single data point will be fed into a custom implementation of the SCluStream algorithm (a real-time clustering machine …

A Review On The Use Of Immersive Technology In Space Research, Mohammad Amin Kuhail, Aymen Zekeria Abdulkerim, Erik Thornquist, Saron Yemane Haile

All Works

Immersive technologies, such as virtual reality (VR), augmented reality (AR), and mixed reality (MR), create digital experiences by merging real and virtual worlds, offering enhanced spatial engagement and sensory immersion. This study examines immersive technologies’ possible advancements to space research, along with application examples in data visualization, astronaut training, and mission planning. Based on the analysis of 44 papers, immersive technologies can assist in diverse areas as varied as procedure guidance, astronaut training, and health-related aspects involving using devices such as HTC Vive, Microsoft HoloLens, and Oculus. The most critical challenges are, by far, difficulties in the selection of participants …

Quantifying The Role Of Active Listening And Reassurance In Virtual Health Coach Interactions, Ghulam Hussain, Brian Keegan, Robert Ross

Conference papers

Conversational Agents have the potential to support healthcare through coaching exercise routines, but are still lacking in demonstrating authentic social behaviours to support engagement. To this end, we present a series of experiments that we conducted in order to investigate how automated health care coaches can be more effective when their interaction style is tailored to demonstrate qualities associated with a good bedside manner, namely active listening and reassurance. To test this, we first developed a dataset of 135 dialogue excerpts from three distinct sources, i.e., original, handcrafted and LLMs, the latter two of which were tuned to demonstrate specific …

Smart Highway Construction Site Monitoring Using Artificial Intelligence, Mehran Mazari, Yahaira Nava-Gonzalez, Ly Jacky N. Nhiayi, Mohamad H. Saleh

Mineta Transportation Institute

Construction is a large sector of the economy and plays a significant role in creating economic growth and national development,and construction of transportation infrastructure is critical. This project developed a method to detect, classify, monitor, and track objects during the construction, maintenance, and rehabilitation of transportation infrastructure by using artificial intelligence and a deep learning approach. This study evaluated the performance of AI and deep learning algorithms to compare their performance in detecting and classifying the equipment in various construction scenes. Our goal was to find the optimized balance between the model capabilities in object detection and memory processing requirements. …

Enhanced Sample Selection With Confidence Tracking: Identifying Correctly Labeled Yet Hard-To-Learn Samples In Noisy Data, Weiran Pan, Wei Wei, Feida Zhu, Yong Deng

Research Collection School Of Computing and Information Systems

We propose a novel sample selection method for image classification in the presence of noisy labels. Existing methods typically consider small-loss samples as correctly labeled. However, some correctly labeled samples are inherently difficult for the model to learn and can exhibit high loss similar to mislabeled samples in the early stages of training. Consequently, setting a threshold on per-sample loss to select correct labels results in a trade-off between precision and recall in sample selection: a lower threshold may miss many correctly labeled hard-to-learn samples (low recall), while a higher threshold may include many mislabeled samples (low precision). To address …

Backdoor Token Unlearning: Exposing And Defending Backdoors In Pretrained Language Models, Peihai Jiang, Xixiang Lyu, Yige Li, Jing Ma

Research Collection School Of Computing and Information Systems

Supervised fine-tuning has become the predominant method for adapting large pretrained models to downstream tasks. However, recent studies have revealed that these models are vulnerable to backdoor attacks, where even a small number of malicious samples can successfully embed backdoor triggers into the model. While most existing defense methods focus on post-training backdoor defense, efficiently defending against backdoor attacks during training phase remains largely unexplored. To address this gap, we propose a novel defense method called Backdoor Token Unlearning (BTU), which proactively detects and neutralizes trigger tokens during the training stage. Our work is based on two key findings: 1) …

Adaptive Deviation Learning For Visual Anomaly Detection With Data Contamination, Aanindya Sundar Das, Guansong Pang, Monowar Bhuyan

Research Collection School Of Computing and Information Systems

Visual anomaly detection targets to detect images that notably differ from normal pattern, and it has found extensive application in identifying defective parts within the manufacturing industry. These anomaly detection paradigms predominantly focus on training detection models using only clean, unlabeled normal samples, assuming an absence of contamination; a condition often unmet in real-world scenarios. The performance of these methods significantly depends on the quality of the data and usually decreases when exposed to noise. We introduce a systematic adaptive method that employs deviation learning to compute anomaly scores end-to-end while addressing data contamination by assigning relative importance to the …