CATA2024: Papers with Abstracts

Abstract. Artificial intelligence (AI) applications are increasingly prevalent in various aspects of our daily lives. One such application is ChatGPT, which has garnered significant interest from professionals across different fields seeking to leverage AI for diverse purposes, in- cluding parallel programming. However, concerns have been raised regarding ChatGPT’s ability to generate correct and efficient code.
The objective of this study is to evaluate the effectiveness of ChatGPT (GPT-3.5), an advanced language model, for parallel programming using OpenACC. Parallel pro- gramming plays a crucial role in accelerating computationally intensive applications and improving overall performance. OpenACC, a parallel programming standard, enables de- velopers to harness the computational power of GPUs for application acceleration.
By conducting a series of experiments, we compare different approaches to parallel programming, with and without the assistance of ChatGPT. Key evaluation factors include execution performance, code quality, and user-friendliness.
The results suggest that integrating ChatGPT can positively impact the parallel pro- gramming process. However, limitations and challenges are identified, such as the need for proper parameter adjustment and reliance on training data utilized by ChatGPT. Sugges- tions for future research directions are provided to address the identified limitations and further enhance the integration of ChatGPT into the parallel programming workflow.

Abstract. The convergence of the Internet of Things (IoT) and blockchain technology represents a pivotal evolution in the digital ecosystem, offering unprecedented opportunities to enhance security, transparency, and efficiency. This paper explores the synergistic potential between IoT and blockchain, aiming to shed light on their dynamic interplay and the prospects it holds for crafting a unified framework. By analyzing the inherent challenges and opportunities within IoT systems and the transformative capabilities of blockchain, we propose a conceptual architecture that could serve as a foundation for future research and development. Our exploration presents a visionary proposal, suggesting pathways for integrating these technologies to realize a robust, scalable, and secure infrastructure for the next generation of IoT applications. This work invites the academic and industrial communities to envision and contribute to the development of innovative solutions that leverage the strengths of both IoT and blockchain, paving the way for a more connected and secure digital future

Abstract. We report in this work the results of our analysis of accuracy of 5 sentiment analysis methods (TextBlob, VADER, logistic regression, support vector machine, CNN on encodings based on BERT tokenization,) for a dataset consisting of tweets from the academia domain, that we API-scraped for 32 universities during the year 2022. We show some results for the volume and sentiment polarity trends exhibited by this dataset. We connect peak and low sentiment averages to concrete events that explain the respective sentiment trend; this proves that observing the social media trends allows to detect real events that need attention and possible action.

Abstract. A rear-end collision happens when a driver collides with the vehicle directly ahead of them from the rear. Such accidents are common at traffic stops like red lights and stop signs or in heavy traffic conditions. While most rear-end accidents occur at low speeds, they can also happen at higher speeds on highways, interstates, and similar fast-moving roadways. Typically, these accidents involve two vehicles, but they can sometimes lead to a domino effect involving multiple vehicles. [1]. This study delves into Mitigating rear-end vehicle collisions using reinforcement learning (RL). The RL algorithm in focus is intended to be integrated into the ego vehicle's system, see Figure 1, aiming primarily to avert colliding with the rear car when both vehicles are progressing forward. Through the utilization of reinforcement learning algorithms, the RCA system can learn from its interactions with the environment, adapt to changing scenarios, and make intelligent decisions to prevent or mitigate collisions effectively. This research investigates the application of the Deep Deterministic Policy Gradient (DDPG) algorithm in the context of rear collision avoidance. The research methodology involves developing a simulated environment that accurately represents lane driving scenarios using longitudinal car dynamics for the ego and rear cars, including the two vehicles’ speeds and positions. The outcomes of this research study are expected to contribute to the development of advanced rear collision avoidance systems that can adapt and improve their performance based on real-time data and experiences.

Abstract. Approximate query processing (AQP) is a computing efficient scheme to provide fast and accurate estimations for big data queries. However, assessing the error of an AQP estimation remains an open challenge for high-dimensional multi-relation data. Existing research often focuses on the online AQP methods which heavily rely on expensive auxil- iary data structures. The contribution of this research is three-fold. First, we develop a new framework employing a non-parametric statistic method, namely bootstrap sampling, towards error assessment for multi-join AQP query estimation. Second, we extend the cur- rent AQP schemes from providing point estimations to range estimations by offering the confidence intervals of a query estimation. Third, a prototype system is implemented to benchmark the proposed framework. The experimental results demonstrate the prototype system generates accurate confidence intervals for various join query estimations.

Abstract. We present a new scheme for storing shortest path information for a polyhedron. This scheme is obtained by applying the constant storage scheme of Han and Saxena [4] on the outward layout of Sharir and Schorr [8]. We achieve constant storage and O(log n + k) time for computing the shortest path from the source point to a query point on the polyhedron, where k is the number of polyhedron edges this shortest path passes through. This improves the result of Chen and Han [3] which uses O(n log n/d) storage and O(d log n/ log d + k) time, where d is an adjustable parameter.

Abstract. Advancements in artificial intelligence continue to impact nearly every aspect of human life by providing integration options that aim to supplement or improve current processes. One industry that continues to benefit from artificial intelligence integration is healthcare. For years now, elements of artificial intelligence have been used to assist in clinical decision making, helping to identify potential health risks at earlier stages, and supplementing precision medicine. An area of healthcare that specifically looks at wearable devices, sensors, phone applications, and other such devices is mobile health (mHealth). These devices are used to aid in health data collection and delivery. This paper aims at addressing the current uses and challenges of artificial intelligence within the mHealth field as well as an overview of current methods to help provide patient data privacy during data collection and storage.

Abstract. In this work, the existing Residue-class based peer-to peer (P2P) network has been considered because of its manifold advantages. Two of the most prominent advantages relevant to the present work are: (1) all peers with the same interest (or possessing same resource type) structurally form a group of diameter one, and (2) the group heads are connected in the form of a ring and the ring always remains connected even in presence of any churn. However, data look-up latency is n/2 for an n group-network. It may become substantial if n is large. To improve the latency, some topological properties of Star inter-connection network have been used to modify the existing RC based network appropriately resulting in remarkable improvement of latency.

Abstract. From the time the Internet was invented to now, security standards have changed dras- tically. Security risks are not only applicable to businesses but also has spilled over into the government and personal realm as well. Newly created threats are being created every day and mitigation techniques need to be capable of being changed in a moment’s notice. Cybersecurity is an environment in which one must adapt to changes or else they will lose. Keyphrases: SHA, RSA, public key, private key, asymmetric key, plaintext, digital sig- nature, security.

Abstract. This paper introduces two distinct new software implementations of ECC over the finite field GF(p) utilizing character arrays and bit sets. Our novel implementations operate on ECC curves of the form y2 = x3 + ax + b (mod p). We have optimized the point addition operation and scalar multiplication on a real SEC (Standards for Efficient Cryptography) ECC curve over a prime field. Furthermore, we have tested and validated the Elliptic Curve Diffie-Hellman key exchange on a real SEC ECC curve using two different implementations of big integer classes. We then proceeded to compare and analyze the performance of these two distinct implementations. Elliptic Curve Cryptography (ECC) represents a promising public-key cryptography system due to its ability to achieve the same level of security as RSA with a significantly smaller key size. ECC stands out for its time efficiency and optimal resource utilization.

Abstract. The importance of document classification has grown significantly in recent years, mostly due to the rise in digital data volumes. Since textual documents often contain more than 80% of all information, there is a perception that text mining has tremendous commercial potential. For future uses, knowledge extraction from these texts is essential. However, it is difficult to obtain this information due to the vast volume of files. As a re- sult, since text classification was introduced, the practice of classifying documents by text analysis has grown in significance. We have primarily employed three different algorithms to compare the metrics between them in order to assess the performance of various models. For this, the dataset was created by extracting condensed information from a variety of textbook genres, including business, social science, and computer science textbooks. To classify textbooks within the same subject group, we used three supervised machine learn- ing techniques in this study: decision trees, random forests, and neural networks. Among these three models, multilayer perceptron neural networks have performed and produced the best outcomes.

Abstract. Robotics is a very expansive and ever-growing field that allows for a large number of different scenarios to occur. Once application is coordination of multiple robots in a single system, and that is highlighted in an environment such as RoboCup competitions. The Contextual Communication and Navigation (CCN) Algorithm is an approach to the problem of multiple robots coordinating at a common goal through the lens of Communication and Navigation. A scheme of communication that allows Nao Robots to efficiently communicate their positions using the minimal number of bits to transfer data was designed as well as a navigation scheme that allows robots to use the information they have on other robots to navigate to a goal. The CCN Algorithm was simulated and tested along with Dijkstra and A* in order to compare the effectiveness of each of the algorithms in tackling this problem and it was found that the CCN Algorithm was the most efficient.

Abstract. Programming infrastructures commonly employ graphs and binary trees to model systems and networks. Efficient operations on trees and graphs are pivotal in enhancing software performance and reducing computational costs, particularly for data-dependent tasks during runtime. This paper analyzes the optimization techniques for recursive algorithms, focusing on the widely used Path Sum algorithm designed for identifying cumulative value sequences that equal a specified target. Employing three distinct techniques—recursion, tabulation, and memoization—this study evaluates their computation time on two prominent data structures: trees and graphs. Results indicate that the memoization approach is completed in less computational time than the regular approach. In contrast, the tabular approach completes in significantly increased computational time, suggesting its inadequacy for traversal optimization. The findings affirm that optimization techniques, particularly memoization, effectively reduce traversal computation time, offering valuable insights for educators and developers working with recursive algorithms in graph and tree-based systems.

Abstract. This work develops the procedure for the construction of a pseudorandom binary generator based in a structure made up of two blocks of four Linear Feedback Shift Register (LFSR) each, which have attached non-linear filtering functions, which deliver binary sequences that are combined by a block that contains the combination devices.
The process includes: characteristics of the LFSR, definition of the model, choice of the different LFSR, selection of Boolean functions based on their optimal cryptographic properties, composition of the generator with the selected elements, key and the procedure to generate them, choice of tests statistics to be used and the criteria for analyzing the results, putting it into operation and carrying out the necessary randomness tests on the sequences obtained.

Abstract. Artificial Intelligence (AI) is and has been rapidly transforming the landscape of cybersecurity as we know it, serving as a double-edged sword. On one side, AI systems can act as your ally, doing tasks like fortifying defense mechanisms and automating complex threat detection or intrusion detection systems. Machine learning models can sift through large amounts of data, identifying anomalies that could indicate a breach or malicious activity, which is often faster and more accurate than a human could be. This capability makes it possible to proactively counteract threats, minimizing any potential cyber issues. On the other hand, the power of AI can be easily weaponized by an adversary, becoming a potential threat actor. Sophisticated cybercriminals can use AI to craft advanced evasive malware and launch automated attacks. Techniques like using machine learning from an adversary perspective allow them to design inputs specifically intended to deceive AI-powered security systems. So as AI systems become smarter, do the tools and tactics used by adversaries looking to exploit them. This duality of AI in cybersecurity is an important relevance to professionals in the field. The discipline of a cybersecurity professional is constantly evolving in response to new strategies of the adversary. Professionals must be skilled at utilizing AI’s capabilities and limiting its weaknesses as it becomes more and more integrated into security infrastructure. Not only must the newest technology be used, but it must also be understood in terms of its ramifications, potential risks, and the constantly shifting dynamics between defenders and attackers. Staying ahead of the competition on the AI-enhanced cyber-warfare battlefield requires constant learning, adaptation, and persistence. In this paper, the discussion pivots around a central question: Is AI in cybersecurity a friend, a foe, or perhaps both?

Abstract. One of the biggest trends in today’s technology and computer science is in the use of natural language processing. Their use in AI has become specifically prevalent in companies such as OpenAI and Google. With their ChatGPT and Bard models, they have made intelligent and social AI models that can mimic human speech and conversation. While talking to these AI models, people can gain vast knowledge by communicating with them. With it being so easy, malicious hackers have started to use it to streamline their attacks. Both companies have tried to put restrictions to help increase the security of their product. However, there are ways to get around it by using different wording that might sound less harmful. This study shows evidence from experimentation with both Google Bard and ChatGPT.

Abstract. Thinking about the experience that the user will have when interacting with a software, is beginning to understand that, in the interaction process, different factors influence: individual, social, cultural, contextual and those specific to the product in question. Artificial intelligence and process automation, among other technologies, are implemented in companies to increase productivity and improve the experience of customers and suppliers. But how do they impact companies? And how can they improve the user experience? Consumers are more likely to purchase brands that they recognize, remember and provide relevant and personalized offers. Personalized experiences can increase conversion rates by 300% and, precisely, Artificial Intelligence (AI) comes to collaborate to achieve this. In this sense, the objective of this proposal is to provide software development companies with a quality model with a people-centered approach that places them at the center of design-driven development. This mode serves as an instrument, guide or good practices that allow them to position themselves at a highly competitive level in the current market, through the production of quality software based on the user experience. For this purpose, a quality model is proposed, the development of which took into account the structure of the International Standard ISO 13407:1999, the ISO 25000 standard and information collected from the software and computer services industry in the region.

Abstract. In this paper, we present a simulation study for a bike-sharing network. The model is analyzed with the Birth-Death process as well as a multidimensional Markov queueing system. We evaluate the steady-state probability of running out of docking stations or bikes. The results provide guidelines for the setup of the network with optimized system efficiency.

Abstract. Large Language Models represent a disruptive technology set to revolutionize the fu- ture of artificial intelligence. While numerous literature reviews and survey articles discuss their benefits and address security and compliance concerns, there remains a shortage of research exploring the implementation life cycle of generative AI systems. This paper addresses this gap by presenting the various phases of the generative AI life cycle and detailing the development of a chatbot designed to address inquiries from prospective stu- dents. Utilizing Google Flan LLM and a question-answering pipeline, we processed user prompts. In addition, we compiled an input file containing domain knowledge of the edu- cation program, which was preprocessed and condensed into vector embeddings using the HuggingFace library. Furthermore, we designed a chat interface for user interaction using Streamlit. The responses generated by the chatbot are both descriptive and contextu- ally pertinent to the prompts, with their quality improving in response to more detailed prompts. However, a significant constraint is the size limit of the input file, given the processing power limitations of CPUs.

Abstract. Cloud usage for storing data and performing operations has gained immense popularity in recent times. However, there are concerns that uploading data to the cloud increases the chances of unauthorized parties accessing it. One way to secure data from unauthorized access is to encrypt it. Even if the data is hacked, the hackers will not be able to retrieve any information from the data without knowing the 'Key' to decrypt it. But when data needs to be used for services such as data analytics, it must be in its original, non-encrypted form. Decrypting the data makes it vulnerable again, which is why Homomorphic Encryption could be the solution to this problem. In this encryption method, the analytical engine can use the encrypted data to perform analysis, where the analysis result will also be in decrypted form. Only authorized users can access the results using the 'Key.' This research proposal proposes a method called pHIDES to enhance data security in the cloud. The pHIDES (pTree-based Homomorphic Intrinsic Data Encryption System) represents data in pTree (Predicate tree) format, a data mining-ready data structure proven to manipulate a large volume of data effectively. The concept of Homomorphic Encryption (HME) along with pHIDES is discussed in our research, along with the algorithmic execution to analyze the effectiveness of the algorithm used to encrypt data in the cloud.