June 22 ~ 23, 2024, Sydney, Australia
Kamronbek Yusupov1, Md Rezanur Islam1, Insu Oh2, Mahdi Sahlabadi2, and Kangbin Yim2, 1Software Convergence, Soonchunhyang University, Asan-si, South Korea, 2Department of Information Security Engineering, Soonchunhyang University, Asan-si, South Korea
This research focuses on evaluating the security of an intrusion detection system in a CAN bus-based vehicle control network. A series of studies were conducted to evaluate the performance of models proposed by previous researchers, testing their effectiveness in real-world scenarios as opposed to those on which they were trained. The article demonstrates that models trained and tested on the same dataset can only sometimes be considered adequate. An approach that included models trained only on CAN ID, Payload, or full data was chosen. The research results show that such methods are ineffective enough in real-world attack scenarios because they cannot distinguish between new scenarios not presented during training. The results of testing the models in various attack scenarios are presented, and their limitations are identified. In addition, a new method is proposed explicitly for attack scenarios that may occur in the real-world use of an in-vehicle CAN communication system.
Intrusion Detection System, Controller Area Network, In-Vehicle Network, LSTM.
Sara Abbaspour Asadollah, School of Innovation, Design and Engineering, Mälardalen University Västerås, Sweden
Ensuring digital privacy is critical to protecting sensitive information and guarding against malicious actors in todays interconnected world. This work-in-progress paper explores the concept of digital privacy and its importance in maintaining online security. We highlight the importance of robust strategies to protect information by examining the consequences of failing to prioritize digital privacy, including identity attack scenarios in industrial applications by proposing a systematic approach to improving digital privacy. Our methodology includes creating a Data Flow Diagram (DFD) to visualize the data flow within the system and applying the STRIDE threat modeling framework to identify potential threats, with a focus on privacy-related aspects. We then extract privacy-related threats and create attack scenarios to guide testers to validate system security. To validate our methodology, we plan to conduct a case study in an industrial application, an automated train control system. By analyzing the data flow and identifying potential attack scenarios, we want to demonstrate the effectiveness of our approach in real-world applications. We also want to automate the process and collaborate with more industries to ensure scalability and practical applicability.
Privacy-related threats, Cyberattack scenario, Information security, Data protection.
Mahmuda Akter and Nour Moustafa, The School of Systems and Computing, University of New South Wales Canberra, ACT 2612, Australia
The Internet of Things (IoT) forms intelligent systems, such as smart cities and factories, to enhance intellectual productivity and provide revolutionary and automated services to end-users and organisations. However, a future IoT ecosystem requires more dynamicity and heterogeneity with advanced privacy preservation. Federated Learning (FL) addresses the challenge of maintaining data privacy by using a privacy-preserving sharing mechanism instead of transmitting raw data. However, the latest cyber threats cause privacy breaches in existing Federated Learning schemes. This study presents a systematic analysis of Federated Learning-based privacypreserving methods in IoT systems. A standard IoT architecture with possible privacy threats is illustrated. Also, Federated Learning schemes and their taxonomies are discussed in a privacypreserving manner, with initial experiments proving the significance of Federated Learning-based privacy preservation in IoT environments and finding acceptable noise addition in differential privacy by keeping higher testing accuracy in different settings to enhance privacy preservation of federated learning. Various Federated Learning schemes, challenges and future research directions are covered.
Federated Learning, Privacy Preserving, Internet of Things (IoT), Privacy Threats.
Carlos Olea, Holly Tucker, Jessica Phelan, Cameron Pattison, Shen Zhang, Maxwell Lieb, Doug Schmidt, Jules White, Department of Computer Science, Vanderbilt University, Nashville, TN, United States
Prompt engineering for the effective use of LLMs has become an important and useful area of research since the advent of highly performant LLMs. Various patterns of prompting have proven effective, among them chain of thought, self consistency and personas. In this paper we measure the effect of both single and multi-agent personas in various knowledge-testing environ- ments, multiple choice and short answer environments. We present our methodology and find that the use of single and multi-agent personas have nominal effects in summation multiple choice tasks and that single-agent expert personas have benefits in short answer tasks. We then hypothesize a spectrum of effect for both single and multi-agent persona usage given our findings.
Prompt Engineering, Large Language Models, Question Answering.
Agampreet Saini,Dr. Rahul Kumar Singh, Dr. Manoj Kumar Sachan, UPES, Dehradun, India
The scope of financial time series prediction has evolved significantly, driven by the surging availability of data and computational advancements. This transition has seen the traditional statistical methods give way to present-day machine learning and deep learning techniques. Despite these strides, challenges continue to exist, with market unpredictability, dimensionality issues, and overfitting in high-frequency data standing as primary obstacles. Newer machine learning models are better than old ones at predicting stocks. When electronic trading and yield optimisation come together, they make these models even better. Deep learning has completely changed machine learning, especially in forecasting financial trends over time. Complex deep learning models lack interpretability, hindering real-world adoption. Considering this we have used two deep learning models which are Long Short-Term Memory and Convolutional Neural Network and trained these on apple stock market data over the years along with one machine learning model i.e. Linear Regression. Then we have compared the accuracy of all the models based on their R-squared Score value.
Financial time series, machine learning, deep learning, market unpredictability, R-squared Score.
Janga Bussaja, Independent Researcher, USA
The system outlined in this proposal exists in a conceptual phase, awaiting the necessary resources for implementation. The theoretical framework presented herein lays the foundation for the development and deployment of Smoky, an innovative artificial intelligence system designed to confront systemic racism. Grounded in African-centered scholarship and equipped with sophisticated monitoring capabilities, Smoky stands as a pioneering endeavor in the realm of leveraging technology for social equity. This scholarly exploration delves into the conceptualization, development, and potential applications of Smoky as a formidable asset in the ongoing struggle against racial injustice. As with any transformative idea, securing funding and support is paramount to transitioning from theory to tangible action. This paper serves as a call to philanthropists and potential collaborators to join in the realization of this vision, contributing to the advancement of technology-driven solutions for social justice.
African-centered Language Model (LLM), Systemic Racism, White Supremacy, Social Justice, Artificial Intelligence (AI) Algorithms, Knowledge Representation, Neural Networks, Data Mining, Machine Learning, Information Retrieval, Natural Language Processing (NLP), Multimedia Analysis, Pattern Recognition, Cognitive Informatics, Planetary Chess, Counter-racist Scholars, Biases in AI, Hybrid Intelligent Systems, Evolutionary Computing, Technical Challenges in AI Development.
Rani Aher, Dr. Mandaar Pande, Department of Symbiosis Centre for Information Technology Symbiosis International University, Plot No:15, Rajiv Gandhi Infotech Park, MIDC, Hinjewadi, Phase 1,Pune, Maharashtra 411057, India
A development of systems capable of compressing audio files is a very appealing research topic. This is due to the necessity to improve storage utilisation and accelerate data transfer over restricted communication channels. As a result, numerous investigators have explored and developed many methods, also to compress audio information with several approaches as well as ways, every one of them have negative consequences such as excessive time consumption or complex computations and remains a key problem. Recent improvements in deep learning has prompted researchers to use unified deep network models to investigate challenges needing highly organised data. The building and design of such models for compressing audio signals has been problematic due to the demand for discrete representations that are difficult to train. This research focuses on quantum neural networks for audio file compression. Our method is based on an innovative encoding process that embeds audio signals in quantum states. For the BBC sound dataset, our framework can compress larger files than previously possible while achieving a best compression ratio than classical neural networks.
Audio compression, Quantum Theory, Quantum Machine Learning, Deep Learning, Quantum States, Q-bit.
Omar Hujran, Department of Statistics and Business Analytics, United Arab Emirates University, Al Ain City, United Arab Emirates
The rapid advancement of smart technologies has significantly bolstered the governments capabilities for digital transformation. The integration of these technologies in various government functions is expanding, showcasing dynamic capabilities that have the potential to fully unlock the transition from digital to data-driven smart government. Despite the widespread acknowledgment of the importance of smart government services, there is a notable dearth of research in this area. Additionally, the existing e-government literature lacks a comprehensive criterion for evaluating smart government, serving as a single reference to guide governments in this transformative journey. Thus, the aim of this study is to fill these gaps in the literature. Results from this study reveal that Abu Dhabi government employs emerging technologies, including artificial intelligence, cloud computing, open-access government data, mobile applications, sensors, and social networking. The significance of the current study lies in the inclusion of specific measures to assess smart government maturity.
Smart Government, E-government, Maturity Model, Digital Transformation, UAE.
Yuanshou Chang1, Yu Sun2, 1Arizona State University, 1151 S Forest Ave, Tempe, AZ 85281, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
In the burgeoning landscape of artificial intelligence, Large Language Models (LLMs) such as GPT have surgedinpopularity, embedding themselves into the fabric of daily digital interactions [1]. As these models assume a pivotal role in shaping discourse, understanding their inherent political biases becomes crucial. This paper delves intothepolitical stance of GPT, examining its consistency and the potential for modification through prompt engineering. Our investigation reveals that GPT exhibits a consistent left-libertarian stance, a finding that underscores theimportance of recognizing and addressing the ideological underpinnings of AI technologies [2]. Furthermore, weexplore the feasibility of altering GPTs political stance towards neutral and right-authoritarian positions throughstrategic prompt design. This research not only illuminates the political dimensions of LLMs but also opens avenuesfor more balanced and controlled AI interactions, of ering insights into the complex interplay between technology, ideology, and user agency.
Prompt Engineering, Artificial Intelligence, Political Bias, Large Language Models (LLMs).
Anurag Yadav1, Neeraj Yadav2, Major Singh Goraya3, 1Sant Longowal Institute of Engineering and Technology, Sangrur, India, 2Bennett University, Greater Noida, India, 3Sant Longowal Institute of Engineering and Technology, Sangrur, India
Selecting a cloud service provider (CSP) from a wide range of options with varying offerings poses a crucial challenge for service requesting consumers (SRCs). The decision criteria outlined in Service Level Agreement (SLA), which defines the Quality of Service (QoS) offered by the CSP, help in selecting a suitable provider. However, relying solely on the SLA for evaluating the merits of a CSP is insufficient. Therefore, the Quality of Experience (QoE) has gained significance in assessing the overall satisfaction of the (SRC). Additionally, monitoring consumer buying behaviour is essential from a business perspective, as all aspects of a business transaction should be mutually beneficial for both the CSP and SRC. Therefore, a new framework is proposed to determine the relative rankings of CSPs and SRCs. Effectiveness of the framework is established through experiments on the QWS cloud dataset, which reveal positive results in terms of implementation complexity and time required.
Cloud Computing, Cloud Service Selection, Quality of Experience, Quality of Service, Multi Criteria Decision making.
Muhammad R Ahmed1, 2, Woshan Srimal1, Thirein Myo1, Mohammed A aseeri3, T. Raja Rani1, Mohammed H Marhaban2, 1Military Technological College, Muscat, Oman, 2Faculty of Engineering, University Putra Malaysia, Selangor, Malaysia, 3King Abdulaziz City for Science and Technology, Riyadh, Saudi Arabia
Computer vision has been increasingly interested in image classification over the past couple of decades. A manual classification of an image can be a very difficult task and time-consuming process; however, automated image classification can provide a highly accurate result using different methods of image classification. It is very challenging to analyze images automatically by a system when compared to the manual classification with human vision. The basic idea behind classification systems is to capture images, and to process those images simultaneously for categorizing them. Basically, image classification involves classifying an image on the basis of similarities and features. If the images are noisy, blurry, have clusters in the background, or are of poor quality, the results may be inaccurate or false. Considering this, it is necessary to develop an accurate algorithm for classifying images. In this work, we have implemented Convolutional Neural Network (CNN) to classify the images. CNN is a deep learning based machine learning algorithm which mimics the human biological vision system and it can automatically detect the features without the human interaction. In future, we would like to combine with method with another algorithm.
Deep Learning, ANN, CNN, Keras, Image Classification.
Vinicius De Souza Santos, Federal Institute of Education, Science, and Technology of S˜ao Paulo (IFSP) - Campus Birigui, Brazil
The global prevalence of diabetes is increasing at an alarming rate, making early and accurate detection a critical area of interest. This study employs Machine Learning techniques to predict the incidence of diabetes in a population of women from the Pima heritage, known for their predisposition to the disease. Using a database of diagnostic measures, multiple algorithms were applied, including Support Vector Machines (SVM), Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Decision Trees, and Random Forest, to develop predictive models. Principal Component Analysis (PCA) was implemented for dimensionality reduction and highlighting of key diagnostic variables, optimizing algorithm performance. The results highlighted the superiority of the Random Forest, which showed higher accuracy and precision, suggesting its viability as a clinical diagnostic tool. This study contributes to the emerging field of artificial intelligence applications in health, providing valuable insights for the prevention and early treatment of diabetes.
Machine Learning, Diabetes, Principal Component Analysis, Random Forest.
Lulu Pi, Xiaoyong Zheng, Roderick M. Rejesus, North Carolina State University, USA
In this study, using experimental data from agricultural experiment stations in Wis- consin from 1990 to 2010 and the recently developed causal forest double machine learning method (CFDML), we estimate the heterogeneous effects of GM corn on yield and yield risk. We find that there is significant heterogeneity in the yield effect of GM corn adoption, but under most growing conditions, the yield effect is positive and sta- tistically significant. Furthermore, we also find significant heterogeneity in the effect of GM corn adoption on yield risk. However, unlike the yield effect, under most growing conditions, the yield risk effect is statistically insignificant. Only under a few scenarios, adopting GM corn has a statistically significant effect, either negative or positive, on yield risk.
GM Corn, Yield and Yield Risk, Causal Forest Double Machine Learning.
Priyanshu Sharma, VIT Bhopal University, India
Wearable medical devices, crafted to monitor and collect health-related information when worn, strive to empower individuals and their healthcare teams by providing insights into health status, facilitating well-informed treatment decisions. The incorporation of artificial intelligence (AI) and machine learning (ML) further enhances these capabilities. AI involves the development and deployment of computer systems capable of executing tasks traditionally associated with human intelligence, such as creating algorithms and models that enable machines to analyze data, engage in reasoning, learn, and make decisions or predictions. Machine learning, a foundational aspect of AI, involves training algorithms on data to recognize patterns and make predictions or classifications without explicit programming. Through iterative learning processes, machine learning models continually improve their performance and adapt to new data. While the potential of AI and machine learning in wearable medical devices is just beginning to be explored, it is evident that these components offer a diverse range of benefits. Nevertheless, it is crucial to acknowledge that they also present specific challenges and considerations that require careful attention.
Machine learning, IoMT, Wearable devices, Monitoring, Healthcare applications.
Forrest McKee and David Noever, 4901-D Corporate Drive, Huntsville, AL, USA
The widespread adoption of voice-activated systems has modified routine human-machine interaction but has also introduced new vulnerabilities. This paper investigates the susceptibility of automatic speech recognition (ASR) algorithms in these systems to interference from near-ultrasonic noise. Building upon prior research that demonstrated the ability of near-ultrasonic frequencies (16 kHz - 22 kHz) to exploit the inherent properties of microelectromechanical systems (MEMS) microphones, our study explores alternative privacy enforcement means using this interference phenomenon. We expose a critical vulnerability in the most common microphones used in modern voice-activated devices, which inadvertently demodulate near-ultrasonic frequencies into the audible spectrum, disrupting the ASR process. Through a systematic analysis of the impact of near-ultrasonic noise on various ASR systems, we demonstrate that this vulnerability is consistent across different devices and under varying conditions, such as broadcast distance and specific phoneme structures. Our findings highlight the need to develop robust countermeasures to protect voice-activated systems from malicious exploitation of this vulnerability. Furthermore, we explore the potential applications of this phenomenon in enhancing privacy by disrupting unauthorized audio recording or eavesdropping. This research underscores the importance of a comprehensive approach to securing voice-activated systems, combining technological innovation, responsible development practices, and informed policy decisions to ensure the privacy and security of users in an increasingly connected world.
Cybersecurity, Voice-activated systems, Automatic speech recognition, Near-ultrasonic frequencies, MEMS microphones, privacy, Acoustic interference, Internet of things, Dgital signal processing, Audio forensics.
Ruoxuan Dong1, Austin Amakye Ansah2, Shuyu Wang3, 1Shanghai Soong Ching Ling School, No.2 Yehui Road, Qingpu District, Shanghai, China, 2The University of Texas at Arlington, 701 S Nedderman Dr, Arlington, TX 76019, 3California State Polytechnic University, Pomona, CA 91768
Most people listen to music to regulate mood, achieve self-awareness, and for socialization [9]. 95.6 percent ofAmericans aged 13 and older listen to some form of audio in their lives. According to a study, people stop findingnew music by age 30. Our solution, Vocodex, provides a way for people to find new music related to their moodthrough emotional state description. Our solution is built with Flutter, which provides the tools needed to build a cross-platform app for Android, Ios, and the web [10]. Our app classifies emotions of up to four different classesand uses that information to fetch songs from a database. Songs are then played with our in-app music player. Themodel trained to analyze text achieved up to 94% accuracy when retrained on a heavier dataset.
Flutter, AI , Music, SVM.
Mujiba Shaima1, Md Nasir Uddin Rana1, Md Tanvir Islam1, Norun Nabi2, Mazharul Islam Tusher1, Estak Ahmed1, Sushanta Saha1, 1Department of Computer Science, Monroe College, New Rochelle, New York, USA, 2Master of Science in Information Technology (MSIT)- Washington University of Science and Technology (WUST), Alexandria, Virginia, USA
IoT, or the Internet of Things, describes a network of networked objects that are equipped with software, sensors, and other technologies to gather and share data. However, blockchain is a distributed ledger technology that makes it possible to record transactions over a network of computers in a safe, transparent, and unchangeable. The way that blockchain and IoT can enhance each others advantages is how they are connected: Blockchain technology, with its decentralized and impenetrable ledger, offers safe and effective storage and transfer of the massive volumes of data generated by Internet of Things devices. Organizations may guarantee the security and integrity of IoT data by incorporating blockchain technology into IoT systems. This will allow for reliable and open communications and transactions between users and devices. Here, we summarize the current body of research and draw attention to the main cybersecurity issues facing blockchain-based Internet of Things platforms. These problems are divided into three primary categories: (i) security of IoT devices; (ii) security of blockchains; and (iii) integration of IoT devices with blockchain (network security). To further address a little about these issues and improve the cybersecurity of blockchain-based IoT systems, we also analysis future research directions.
IoT, Blockchain, Blockchain Layer, Hash Identification.
Zerihun Olana Asefa and Admas Abtew, Jimma University, Department of Information Technology
News recommendation schemes utilize features of the news itself and information about users to suggest and recommend relevant news items to the users towards the interest they have. However, the effectiveness of the existing news recommendation scheme is limited in the occurrence of new user cold start problems. Therefore, we designed a news recommender system using hybrid approaches to address new user cold start problems to ease and suggest more related news articles for new users. To achieve the objective mentioned above, user demographic data with a hybrid recommendation system that contains the scheme of both content-based and collaborative filtering approaches is proposed. To evaluate the effectiveness of the proposed model, an extensive experiment is conducted using a dataset of news articles with user rating value and user demographic data. The performance of the proposed model is done by two ways of experiment. So, the performance of the proposed model performs around 68.05% of Precision, 42.46% of Recall and 52.1% of the average F1 score for the experiment based on individual user similarity in the system. And also performs around 93.75% of precision, 40.25% of recall and 56.31% F1-score for the similarity of users based on the similarity of users within the same category which is better than the first experiment.
news recommendation system; cold start problem; hybrid approach; demographic information; new users; popular news.