Accepted Papers

Comparing a Composite Model Versus Chained Models to Locate a Nearestnsec Visual Object

Antoine Le Borgne1, Xavier Marjou1, Fanny Parzysz1 and Tayeb Lemlouma2, 1Orange Innovation, Lannion, France, 2IRISA, Lannion, France

ABSTRACT

Extracting information from geographic images and text is crucial for autonomous vehicles to determine in advance the best cell stations to connect to along their future path. Multiple artificial neural network models can address this challenge; however, there is no definitive guidance on the selection of an appropriate model for such use cases. Therefore, we experimented two architectures to solve such a task: a first architecture with chained models where each model in the chain addresses a sub-task of the task; and a second architecture with a single model that addresses the whole task. Our results showed that these two architectures achieved the same level performance with a root mean square error (RMSE) of 0.055 ± 0.120 and 0.056 ± 0.113; The findings further revealed that when the task can be decomposed into subtasks, the chain architecture exhibits a twelve-fold increase in training speed compared to the composite model. Nevertheless, the composite model significantly alleviates the burden of data labeling.

KEYWORDS

Chained models, Composite model, Transformer, CNN, GPT, GPT-4.

Forest Mixing: Investigating the Impact of Multiple Search Trees and a Shared Refinements Pool on Ontology Learning

Marco-Silviu Pop-Mihali and Adrian Groza, Technical University of Cluj-Napoca, Romania

ABSTRACT

We aim at development white-box machine learning algorithms. We focus here on algorithms for learning axioms in description logic. We extend the Class Expression Learning for Ontology Engineering (CELOE) algorithm contained in the DL-Learner tool. The approach uses multiple search trees and a shared pool of refinements in order to split the search space in smaller subspaces. We introduce the conjunction operation of best class expressions from each tree, keeping the results which give the most information. The aim is to foster exploration from a diverse set of starting classes and to streamline the process of finding class expressions in ontologies. The current implementation and settings indicated that the Forest Mixing approach did not outperform the traditional CELOE. Despite these results, the conceptual proposal brought forward by this approach may stimulate future improvements in class expression finding in ontologies..

KEYWORDS

Ontology Learning, DL-Learner, Inductive Logic Programming (IDL), Description Logic (DL), White-box Machine Learning.

Advanced Uncertainty Quantification and Novelty Detection for Random Forest Models

Janne Merilinna, VTT Technical Research Centre of Finland, Espoo, Finland

ABSTRACT

In practical applications, model accuracy alone is insufficient; quantifying model uncertainty is crucial, particularly in mission-critical scenarios involving life, money, or reputation. In this paper, we propose a novel method called MACAU (Model-based AleatoriC and epistemic uncertAinty qUantification) and implement it in the LightGBM gradient-boosting framework. MACAU enables the quantification of both aleatoric and epistemic uncertainties in Random Forest (RF). Additionally, MACAU offers enhanced novelty detection capabilities, particularly valuable for identifying out-of-distribution (OOD) samples. We compare MACAU with other RF- or gradient boosted trees-based methods, including RF-native between-variance, quantile regression, inductive conformal prediction, exogeneous model for uncertainty estimation using the Gaussian negative log-likelihood method, Natural Gradient Boosting, and CatBoost. Our evaluation is conducted on both synthetic and real-world regression cases. The results demonstrate the effectiveness of MACAU in quantifying model uncertainty, as measured by the Continuous Ranked Probability Score, as well as detecting OOD samples, as measured by the ROCAUC.

KEYWORDS

Machine Learning, Epistemic and Aleatoric Uncertainty, Out-of-Distribution Detection.

Find Drivable Segments from Road Image using Depth and RGB Image-Final Version

Xuemei Li, Department of Computer Engineering and Engineering, Oakland University, Rochester, USA

ABSTRACT

Perception is a very critical and challenging task in the realm of autonomous driving. The current approach relies on a sophisticated model pipeline built upon various deep learning models, each tasked with solving distinct challenges. This leads to a large model size. The proposed approach dissects an entire driving scene into two distinct elements: driving backgrounds that are not for vehicles to drive onto and road segments that are for vehicles driving. It performs the pointwise fusion using disparity image and RGB image. It uses pointwise and depthwise convolution to reduce multiplication times. It integrates image segmentation neural networks Deeplab V3 as backbone and significantly reduces the model size using ResNet-18. The efficacy of the proposed neural network is substantiated through validation using the Cityscape dataset, yielding an impressive 0.979 accuracy, 0.948 precision, and a 0.947 F1-score. Furthermore, it boasts a training speed that is five times faster compared to conventional UNet- based models, and the model size is eight times smaller than UNet-based models.

KEYWORDS

Image Segmentation, Image Perception, Object Detection, UNet, Deeplab V3, ResNet-18.

Automate the Generation of the Flowchart Visualization Using Artificial Intelligence and Static Analysis

Qianzhi Li1, Marisabel Chang2, 1Shenzhen College of International Education, Antuoshan 6th Road No.3, Futian District, Shenzhen, Guangdong, China, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768

ABSTRACT

In todays evolving software landscape, the visualization tools for understanding complex systems must be both adaptive and advanced [4]. Traditional flowcharts, although essential, are manual and prone to inaccuracies when used for depicting intricate programs. This paper presents a methodology to automate flowchart visualizations using the synergy of Artificial Intelligence (AI) and Static Analysis, paying meticulous attention to code constructs and vital keywords. Our system is adeptly engineered to receive code in a .txt format and, post-analysis, delivers a .dot file, vertices.txt file, and an illustrative control flow graph (CFG) [5]. It boasts an intrinsic capability to identify and differentiate common control conditions such as if, else, while, for, and do. Moreover, its fine-tuned to recognize and process essential programming keywords like break and continue, ensuring the visualizations mirror the actual behavior of the code. A pivotal element of this methodology is the Node class, designed to encapsulate and manage crucial node information [6]. This class guarantees that every vertex in the CFG carries its inherent details, fostering a rich, in-depth understanding of the codes logic and flow. The system employs a Lexer to tokenize the code, forming the foundational layer for the subsequent steps of analysis and visualization. Recognizing the importance of accessibility and usability, weve taken our solution online. We built a website utilizing HTML, CSS, and JS, creating a user-friendly interface where users can effortlessly input their code and generate corresponding flowcharts on-the-fly. This web interface not only democratizes access but also facilitates real-time interaction and modifications. Through extensive evaluations, our approach showcased commendable accuracy. This paper highlights the transformative potential of our method in revolutionizing code documentation, debugging processes, and educational tools, bridging the divide between complex code structures and their lucid visual counterparts [7].

KEYWORDS

Static Analysis, Flowchart, Abstract Syntax Tree.

A Personalized Mental Health Application to Enhance Well-being Using Artificial Intelligence and Usercentered Design

Calvin King1, Ruiyan Fang1, Yujia Zhang3, 1Pacific academy, 4947 Alton Pkwy, Irvine, CA 92604, 2Army and Navy Academy, 2605 Carlsbad Blvd, Carlsbad, CA 92008, 3Computer Science Department, California State Polytechnic University, Pomona, CA 91768

ABSTRACT

This paper presents a solution to address the pressing societal challenges associated with mental health support, particularly among the elderly and individuals facing conditions such as anxiety, depression, and PTSD [4]. The rapidly aging global population and the increasing prevalence of mental illnesses underscore the need for accessible and effective interventions [5]. We propose a personalized mental health app that leverages artificial intelligence to curate customized playlists aimed at alleviating the negative effects of these mental health conditions. This app prioritizes accessibility, affordability, and user satisfaction while aiming to bridge the gap in mental healthcare. Key technologies and components of the app include AI-driven playlist customization, user-friendly interfaces for diverse user demographics, and data privacy measures [6]. Challenges such as accuracy in playlist tailoring and usability for elderly users were addressed through iterative refinement. Experimental results demonstrated the apps effectiveness in improving mental well-being and user satisfaction, validating its potential as a valuable tool in mental health support. We believe this idea offers a transformative solution that people should embrace, as it combines technologys power with empathy to provide accessible and personalized mental health care, meeting the critical needs of underserved populations [7].

KEYWORDS

Mental Health, Artificial Intelligence, Healthcare, Mental Health Interventions.

Study of Video Object Motion Tracking Using Dense Optical Flow Techniques

Hari K.C.1, Manish Pokharel2, Sushil Shrestha2, 1Department of Electronics and Computer Engineering, Tribhuvan University, Nepal, 2Department of Computer Science and Engineering,Kathmandu University, Nepal

ABSTRACT

Digital video content refers to video content that is stored and transmitted in a digital format. Video contains the Objects and its motion. Object detection and motion tracking are two of the most fundamental tasks in digital video content analysis. The main goal of this paper is to detect the multiple objects and track the motion in dynamic environment. This paper provides motion tracking of multiple objects using dense optical flow approaches such as FarneBack method and FarneBack-Gaussian method. Video dataset used for the experiment are collected from YouTube. The experiment result shows that FarneBack-Gaussian dense optical flow approach provides higher precision, stability and computationally efficient than FarneBack dense optical flow. Various mechanisms like F-score and time complexity will be used to test and validate results. The significance of this research is to extract the information about the motion and behavior of objects within video in dynamic environment. Security companies, sports organizations, autonomous vehicle manufacturer, government agencies, academic researchers will be benefitted from this research.

KEYWORDS

Object detection, motion tracking, dense optical flow, Farneback method, Farneback-Gaussian method

Machine Learning Methods for Predicting Exam Results

Imran Chamieh, Klaus Giebermann and Christian Weiß, Department of Natural Science, Ruhr West University of Applied, Sciences, Mülheim an der Ruhr, Germany

ABSTRACT

In improving the learning processes, predicting students performance becomes an essential factor. If we can detect at-risk students, i.e., those who might not pass the exam, early enough, we can reduce the drop-out rate and enhance students performance. In this paper, we build various machine learning ML and deep learning DL models to predict student performance by analysing the log data for online exercises generated at the end of the class phase. The models we have created are specified for one course (Mathematics I). This course is supported by mandatory online exercises, that the students have to work on during the semester. Our study introduces two models. The first aims to identify students who may not pass the examination, while the second predicts students final exam grades, rounded to the nearest integer. Through applying various machine learning and deep learning algorithms, we have achieved encouraging accuracy in predicting pass or fail outcomes. Predicting precise exam marks proves more challenging, given the variance in student studying behaviours and their corresponding results. To account for this, we propose a new, more lenient accuracy metric. This metric deems a prediction correct if it matches the actual mark or is within one mark of the actual score. With this metric, our models demonstrate promising prediction accuracy, with the deep learning model (Convolutional Neural Network CNN) showing significant potential for effectiveness.

KEYWORDS

Online Learning; Learning Analytics; LMS; Log Data; Student Success; Educational Data Mining (EDM).

A Transformative Program to Empower Female Students in Stemusing Ai Chatbot Technology

YanyanZhou1, MarisabelChang2, 1ChongqingNankai Secondary School, No.1 Shanan Street, Shapingba, Chongqing400030, 2computer Science Department, California State Polytechnic University, Pomona, CA91768

ABSTRACT

Inthe realmofSTEMeducation,theunderrepresentationoffemalestudentsremainsapersistentchallenge [4].This research paper addresses this issue by proposing an innovative educational platform empowered by AI and chatbot technology [5]. The background problem revolves around the multifaceted challenges faced by female students in STEM, including a lack of mentors, self-doubt, and gender biases. To tackle this problem, our proposal centers on the development of an educational platform designed to offer personalized learning resources and an AI chatbot mentor [6]. Key technologies and components include AI-driven chatbots, personalized content curation, and realtime mentorship features. Challenges encountered during the project included ensuring the effectiveness of the AI chatbot mentor and curating diverse and relevant learning resources. We applied our platform to various scenarios during experimentation, involving female students in STEM fields [7]. The results indicated positive feedback and increased confidence among users, emphasizing the potential of our idea in addressing gender disparity in STEM. Ultimately, our innovative platform offers an accessible, inclusive, and effective solution to empower female studentsin STEMandpromote diversity in these critical fields, makingita valuable tool foreducational institutions andaspiringstudentsalike.

KEYWORDS

STEMEducation,GenderDisparity,AIChatbot,PersonalizedLearning.

Comparing a Composite Model Versus Chained Models to Locate a Nearest Visual Object

Antoine Le Borgne1, Xavier Marjou1, Fanny Parzysz1 and Tayeb Lemlouma2, 1Orange Innovation, Lannion, France, 2IRISA, Lannion, France

ABSTRACT

Extracting information from geographic images and text is crucial for autonomous vehicles to determine in advance the best cell stations to connect to along their future path. Multiple artificial neural network models can address this challenge; however, there is no definitive guidance on the selection of an appropriate model for such use cases. Therefore, we experimented two architectures to solve such a task: a first architecture with chained models where each model in the chain addresses a sub-task of the task; and a second architecture with a single model that addresses the whole task. Our results showed that these two architectures achieved the same level performance with a root mean square error (RMSE) of 0.055 ± 0.120 and 0.056 ± 0.113; The findings further revealed that when the task can be decomposed into subtasks, the chain architecture exhibits a twelve-fold increase in training speed compared to the composite model. Nevertheless, the composite model significantly alleviates the burden of data labeling.

KEYWORDS

Chained models, Composite model, Transformer, CNN, GPT, GPT-4.

Improve Navigation of the Webmail Interface Using Arabic Voice Commands for Elderly and Disabled Employees: a Computer Application

Mokhtar Alkhattali1, Mostafa Dow2 and Khawla Azwee3, 1Department of Computer Science, High Institute of Science and Technology, Qaser Bin Ghashir, Libya, 2Department of Computer Science, College of Science and Technology, Jadu, Libya, 3Department of Computer Science, High Institute of Science and Technology, Qaser Bin Ghashir, Libya

ABSTRACT

The development of humanitarian assistance applications has revolutionized business efficiency and daily convenience. Voice recognition technology (VRT), with its improved accuracy, has found extensive use in various fields, including assistance programs for individuals with disabilities or limited mobility in vehicles, homes, and websites. The authors have developed "Asis_Webmail," a computer application (PCApp) written in Python, to enhance the accessibility and usability of Webmail for Arabic-speaking seniors and physically disabled employees. The application allows users to navigate the webmail interface using Arabic Voice Commands (AVC), promoting independence and functionality in their daily lives. The effectiveness of "Asis_Webmail" was assessed through a survey of disabled employees, who reported finding the application useful and mentioning improved interaction with their email interface. Ultimately, this application aims to empower Arabic-speaking individuals, regardless of mobility disability levels, to independently use the Webmail interface using AVC, thereby promoting independence in both social and functional aspects of their lives.

KEYWORDS

Voice Recognition Technology, Webmail, Older Employees, Assist Applications, Arabic Voice Commands.

Solving the Knapsack Problem Using a Cultural Algorithm

Mohammad Saleh Vahdatpour, Department of Computer Science, College of Arts & Sciences, Georgia State University, Atlanta, USA

ABSTRACT

Cultural Algorithms (CA) are metaheuristic optimization algorithms. In this paper, we propose a variation of CA for solving 0-1 knapsack problems. The proposed algorithm considers a belief space to sift population and introduces two functions for adjusting the rate of crossover and mutation operators in the evolutionary process. Experimental results demonstrate the efficiency of the proposed algorithms in finding the global optimum for knapsack problems with high dimension.

KEYWORDS

knapsack problem, cultural algorithm, genetic algorithm.

Application of a Risk Focused Simulation Based on a V&v Concept for Auto-route Generation for Seagoing Vessels

Arnold Akkermann1, Bjorn Age Hjollo2, 1German Aerospace Center,Institute of Systems Engineering for Future Mobility, Department Application and Evaluation Oldenburg, Germany, 2NAVTOR AS Egersund, NORWAY

ABSTRACT

Route design is one of the main tasks of a navigator. The quality of the route is related to the safety of navigation and the economic benefits for ships. In general, routes should be able to meet two main objectives, namely avoiding dangerous areas and ensuring economic benefits. For maritime transport, it is important to be able to design safe routes quickly and easily. Route planning is usually done by the captain and chief officer. Although this approach does not cause problems in most cases, the manually designed route is still influenced by the professional quality, sailing experience and personal emotions of the crew. It is therefore not necessarily the optimum route. With the development of computer technology, and information technology researchers began to use heuristic algorithms to automatically generate ship route and many methods and procedures have been developed to determine the best route available. Within this paper we present our approach to the verification and validation of auto-routes for seagoing vessels using two virtually generated reference routes.

KEYWORDS

Generation of comprehensive worldwide routes for seagoing vessels, AIS data, Reference-Routes, Verification and Validation in virtual world.

Chinese Named Entity Recognition Based on Knowledge-based Question

ArvindChandrasekaran,Colorado Technical University,United States of America

ABSTRACT

The knowledge-Based Question Answering (KBQA) system is an essential part of the customer service system aiming to answer natural language questions by recovering the structured data stored under the knowledge base. KBQA answers the natural language questions by recovering the structured data stored under the knowledge base. KBQA receives the user’s query and first needs to recognize the topic for the query entities like the location, name, organization, etc., The process is Named Entity Recognition (NER) using the Bidirectional Long Short-Term Memory Conditional Random Field model, and the SoftLexicon method is introduced as the Chinese NER tasks. A fuzzy matching module is proposed to analyze the application scenario characteristics using multiple methods. The module efficiently modifies the error recognition results, improving entity recognition performance. The fuzzy matching and the NER model are combined into the NER system. The power grid-related original data is collected to improvise the system following the power grid data characteristics.

KEYWORDS

Knowledge-Based Question Answering; SoftLexicon method; Knowledge graph learning representation; System Of Question answering; Knowledge; Spatial Temporal.

Educational Applications on Smartphone for Virtual Laboratories in Engineering

Anthony Khoury1, Rafic Younes2, Bilal Hoteit3, Said Abboudi4, Jihan Khoder5, Rachid Outbib6, 1Aix-Marseille University, France, 2Lebanese University, Lebanon, 3Lebanese University, Lebanon, 4University of Technology of Belfort-Montbéliard, France, 5ECE School Of Engineering, 6Aix-Marseille University, France

ABSTRACT

Many universities provide remote and online teaching for several majors. The importance of online teaching has increased and it has become very popular. One of the main problems that have emerged is how to conduct the scientific experiments. This paper focuses on the transition of virtual lab to the smartphone platform. It presents an overview of laboratory types then the advantages and disadvantages of virtual labs are discussed. Afterwards, a set of challenges for the development of virtual labs is presented. Solutions for these challenges are then presented alongside their implementation in a developed virtual lab application. In particular, the virtual lab applications employed are concrete mix design, fluid friction test, machine learning lab, control simulation andpermeability testing.

KEYWORDS

Virtual lab, mix concrete design, fluid friction, machine learning, elevator control, permeability test, smartphone application.

Emotional Protection System for Emotional Laborers

CHANG-MOOK OH, HelpU, Inc., Seoul, South Korea

ABSTRACT

This paper proposes and develops LINCALL, a real-time voice processing and machine learning system designed to provide preemptive emotional protection for emotional laborers working in environments such as call centers. LINCALL analyzes voice data in real-time conversations with customers to understand their emotional states, converting this data into text for analysis. The system identifies keywords and offers appropriate questions and answers, aiding emotional laborers in handling customer interactions more effectively. It also includes a feature for responding appropriately to the customers emotional state, such as voice blocking. Additionally, the automatic voice-to-text conversion feature reduces fatigue for emotional laborers, enhances the efficiency of consultations, and aids in preemptive emotional protection for emotional laborers.

KEYWORDS

Emotional laborers, Emotion analysis, Speech recognition, Pre-emptive emotional protection, Machine learning.

Design Proposal of a Virtual Platform Using Artificial Intelligence to Improve the Quality of Mathematics Learning

Gutierrez Alberto1 and Aguilar-Alonso Igor2, 1Department of Computer Engineering, San Marcos University, Lima, Cyprus, 2Systems Engineering School, National Technological University of South Lima, Lima,Perú

ABSTRACT

There are various factors that are determinant in learning mathematics at school, from difficulties associated with learning mathematics to the study environment. The present study analyzes the possible factors that positively or negatively affect the learning of students in the school stage, as well as the methodologies that are used and their relationship with technological tools today, in order to improve the quality of the teaching of mathematics. Thus, a proposal for the design of a virtual platform is also presented whose purpose is to improve the quality of mathematics teaching using tools based on artificial intelligence such as recommender systems, intelligent tutors, text analysis systems, among others.

KEYWORDS

Learning, mathematics. Artificial intelligence, schools.

Training a Yolov5 Model on a Custom Dataset Using a Gpu Nvidia A100-pcie-40gb

Maria Laura Clemente, CRS4, Italy

ABSTRACT

The present work describes the results of a Yolov5 model training on a Dell server R750 with a GPU NVIDIA A100-PCle-40GB (called A100 here after), compared with similar trainings on a HP Workstation Z1 G8 with a GPU NVIDIA GeForce RTX 3070 (called RTX3070 here after). The aim of the activity was related to the use of a Yolov5 model for object detection, based on a custom dataset, made of 5579 images belonging to 12 similar categories.

KEYWORDS

Object Detection Performances, Deep Learning, Yolov5, model training.

Lightweight Certificateless Authenticated Key Agreement Protocol

Dr. Eng. Mwitende Gervais, Pivot Access Ltd, Kigali, Rwanda

ABSTRACT

Data security and privacy are important to prevent the reveal, modification and unauthorized usage of sensitive information. The introduction of using critical power devices for internet of things (IoTs), e-commerce, e-payment, and wireless sensor networks (WSNs) has brought a new challenge of security due to the low computation capability of sensors. Therefore, the lightweight authenticated key agreement protocols are important to protect their security and privacy. Several researches have been published about authenticated key agreement. However, there is a need of lightweight schemes that can fit with critical capability devices. Addition to that, a malicious key generation center (KGC) can become a threat to watch other users, i.e impersonate user by causing the key escrow problem. Therefore, we propose a lightweight certificateless Authenticated Key Agreement (AKA) based on the computation Diffie-Hellman problem (CDHP). The proposed protocol maintains the characteristics of certificateless public key cryptography. The protocol is split into two combined phases. In the first phase, our protocol establishes a session key between users (sender and receiver). In the second phase, we use a lightweight proxy blind signature based on elliptic curve discrete logarithm problem (ECDLP). The used proxy signature has small computation costs, and can fit for small devices such sensors and protects against un-authentication and un-authorization on decentralized system. Compared to the existing AKA schemes, our scheme has small computation costs. The protocol achieves the well known security features compared to the related protocols.

KEYWORDS

Cerificateless AKA · distinguishability· Session key · proxy blind signature · forward secrecy · decentralized.

The Role of Personality in Character Animation

Nicholas Bruggner Grassi, Bruna Goddini de Oliveira Ferreira, Eduarda Dippe Ramos Rondon and Morgana de Fraceschi Hoefel, Department of Management, Media and Tecnology, Universidade Federal de Santa Catarina, Florianópolis, Brazil

ABSTRACT

This text refers to the development of an animation style for the main character of the short film "Home Sweet Home." It addresses the importance of producing movement tests during the process of constructing a characters personality, thus introducing the animators role in a collaborative process with various professionals involved in the pre-production of a project. As methodological tools, literature research was conducted on the process of constructing a characters personality and how it relates to his animation. Based on this study, the character animation for a teaser was developed with the intention for it to serve as a guideline for the subsequent production of the short film.

KEYWORDS

Character animation, Character personality, Early character movement test.

The Role of Personality in Character Animation

Hocine SAADI1 and Malika Mehdi2, 1CERIST Research Center, Algiers, Algeria, 2Universite des Sciences et de la Technologie Houari Boumediene, Algiers, Algeria

ABSTRACT

Virtual screening (VS) is a computational technique used in the pharmaceutical industry in the process of discovering new drugs. The aim is to identify structures which are most likely to bind to a drug target by screening a library of small molecules. Since this problem is complex and known to be NP-hard, several metaheuristics have been proposed to solve it. The scoring function of those metaheuristics is the bottleneck and needs a huge computing time. Nowadays, such computation power can be provided by High-Performance Computing (HPC) such as Graphics Processing units (GPUs), Many-Core clusters and Multi-core CPUs. This paper presents parallelization techniques we propose to speed up the scoring function calculation, and a corresponding evaluation in terms of computation time achieved on different HPC architectures.

KEYWORDS

HPC, Virtual screening, Molecular Docking, Scoring function, GPU, Many-Core, Multi-core.

A User-friendly Mobile Application to Facilitate Art Commissions Using Innovative Ui/ux and Data Management Technologies

Fangyijng Huang1, Ang Li22, 1Sage Hill School, 20402 Newport Coast Dr, Newport Beach, CA 92657, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768

ABSTRACT

This paper addresses the challenge of connecting artists and individuals seeking commissioned art by introducing a dedicated mobile application [4]. Artists often rely on commissions for income, yet theres a lack of efficient platforms for facilitating these connections. Our proposed solution involves a user-friendly iOS and Android app with features like account creation, artist discovery, commissioning tools, and a Dropdown feature for setting deadlines. Challenges during development included UI/UX design complexities and limited user feedback, which were tackled through iterative improvements and simulated user experiments [5].Usability and user satisfaction assessments revealed strong positive feedback for user interface design and the commissioning process, while pointing out areas for enhancement in the Dropdown and account creation [6]. Ultimately, this app serves as a transformative solution, fostering a vibrant art community and providing an efficient, user-friendly platform for commissioned art projects [15]. It addresses real-world challenges for artists and art enthusiasts and offers a valuable tool to enhance creative collaborations, making it a compelling solution for users seeking commissioned artwork.

KEYWORDS

Art, Commission, Social, FlutterFlow.

Web Based Content Delivery Using Learning Management Systems Under Higher Education

Arvind Chandrasekaran, Colorado Technical University

ABSTRACT

Information technology advances facilitate new training and educational practices, thereby creating opportunities for new tools and methods, and these technologies modify educational patterns. The intelligent education frameworks are mostly qualitative studies, thereby enumerating the conceptual framework of the education systems smartly. Peer and teacher interaction and the teaching practice accommodate comprehending the students’ status and need in multiple ways and thereby provide the students with real-time asynchronous/synchronous help and guidance. The experts surveyed by the Analytic Hierarchy Process (AHP) were selected from E-learning and information technology totaling 14 people. The sub-criteria weight values display the E-Learning Education Planning key factors, which are the optimizations of educational resources; teaching is differentiated among the person, rate of course completion, and wireless connection reliably. The minor importance is the innovative pedagogies based on learning attention, theory, and data backup. Regarding contribution, only some studies used hierarchical design, and the AHP model presents the current research.

KEYWORDS

Smart classroom; Smart education design; Smart e-learning education.

Authentication in Atm/itm Machines Using Iris Recognition Biometrics

Hafızullah ozgur and Yunus Emre Selcuk, Department of Computer Engineering, Yıldız Technical University, Istanbul, Turkey

ABSTRACT

Abstract—Biometric identification is a method that encompasses many types of authentication systems, including iris features and characteristics. The behaviour of the tiny pixels in a person’s iris and pupil can be extracted and uniquely used for authentication in ATM/ITM machines. In this study, achieving higher accuracy and success is the goal. We have found out that the Daugman algorithm can enable the implementation of iris biometrics in a faster and more accurate way compared to other algorithms. This will allow ATM/ITM users can comfortably access their accounts and make transactions without the need for a card or PIN, while authenticating their identity in the machines using their iris biometrics. Such a system can also be used to authenticate national ID entitlement in programmes. Furthermore, such a system can enhance the accessibility of social benefits, subsidies, and other entitlements while minimizing fraudulent activities [1].

KEYWORDS

Iris Recognition, biometric authentication, Daugman algorithm, ATM Authentication by iris biometrics.

Real-time Detection of Children in the Front Seat of a Car Using Deep Learning Algorithms

ABSTRACT

The use of automated technology has the potential to enhance road safety and prevent fatalities. A new approach for detecting the presence of a child in a cars front seat using image processing techniques can prevent accidents caused by children being in the front seat. By analyzing images from a surveillance camera, we can identify the location and analyze pixel values using machine learning algorithms to determine if a child is present in the image. This technology can help prevent severe accidents caused by leaving children in the front seat of a car. Our goal is to develop an AI-based deep learning algorithm that can detect children in the front seat of a car by analyzing various features of the child and classifying them as either a child or adult. This algorithm will also be used to detect the presence of people in the car and monitor the street, automatically sending traffic violation information to the driver. Our training dataset had 2,624 images, the validation dataset had 254 images, and the testing dataset had 316 images. We achieved a training accuracy of 97%, a validation accuracy of 95.67%, and a testing accuracy of 86%. The classification report is displayed in the figure provided.

KEYWORDS

Children detection, front seat, image processing, car, deep learning.

A Cost-effective Virtual Sensor for Continuousfreshwater Nutrient Monitoring Using machinelearning

Andrew Zhou1, Ivan Revilla2, 1Lexington High School, 251 Waltham St, Lexington, MA 02421, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768

ABSTRACT

Nutrient enrichment of aquatic environments is a prevalent issue with wide-reaching negative implications forecological stability, tourism and recreation, and vital drinking supplies. Proper management of nutrient influxes—primarily nitrogen and phosphorus—into aquatic environments is facilitated by continuous monitoring of nutrient levels within water bodies of interest, which of ers a more complete understanding of seasonal trends and fasterresponse times compared to traditional lab testing. However, continuous nutrient monitoring systems areprohibitively expensive, with ongoing energy and maintenance requirements that limit deployment. Machinelearning shows potential for virtual sensor development with real-time nutrient prediction, based on continuouslymonitored surrogate indicators. In this study, we test the feasibility of this premise by evaluating the performanceof Random Forest regressor (RF), k-Nearest Neighbors (kNN), Support Vector Machine regression (SVM), DecisionTree regressor, Artificial Neural Network, Gradient Boosting Regressor (GBR), and Histogram Gradient BoostingRegressor (HGBR) on one year of water quality testing data from sites across the Continental United States(CONUS). To address values missing not at random, an issue prevalent in water quality testing data, important surrogate indicators are identified by permutation importance. Models are then trained and tuned with BayesianOptimization to identify hyperparameters optimal for explaining target variance. Across both phosphorus andnitrogen prediction, RF achieved the highest validation performance, with GBR and HGBR trailing marginally. These findings suggest that ensemble tree models are well-suited to continuous nutrient monitoring and may beacost-ef icient solution to greatly supplement the existing high-frequency testing network.

KEYWORDS

Freshwater, Total Phosphorus, Total Nitrogen, Ensemble Tree, Bayesian Optimization.

A Feature Selection Approach to Identify Key Performance Indicators in Simulated Racing

Fazilat Hojaji, Adam J. Toth, John M. Joyce, Mark J. Campbell, Esports Science Research Lab, Lero, University of Limerick, Limerick, Ireland

ABSTRACT

Feature selection, referring to the reduction of input variables to develop the best performing predictive models, is a critical step in data analysis and machine learning. Sometimes, learning algorithms do not perform optimally when data are applied directly for classification or clustering purposes. This could be due to the existence of irrelevant, redundant, and/or ‘noisy’ features. To overcome this challenge, the current article proposes a hybrid feature selection approach with three steps; filtering, wrapping, and interpreting. This approach is subsequently used to predict the performance of simulation (sim) racing in electronic sports (esports). A dataset was obtained from 93 sim racing drivers using a racing simulator, and our proposed hybrid feature selection approach was applied. We found that the resulting model produced by our hybrid approach (86.5%) outperformed wrapper-only methods (73.89%) in terms of accuracy in predicting lap time. Furthermore, the experimental results also show that our approach can accelerate learning, increase quality, and detect important features. Our proposed feature selection approach has applicability spanning beyond esports.

KEYWORDS

Feature selection, Sim racing, Telemetry data, Classification.

Accurate Dns Server Fingerprinting Based on Borderline Behaviour Analysis

Botao Zhang, College of Electronic Engineering, National University of Defense and Technology, Hefei, Anhui, China

ABSTRACT

Server fingerprinting is widely employed as pre-exploit technique by attackers as well as security researchers. Recent security incidents against global DNS ecosystem show close association with outdated versions of DNS servers. In this paper, we demonstrate malicious actors are still able to infer the DNS server fingerprints by analyzing the borderline behaviour of different DNS servers, despite administrators have hidden DNS server fingerprints by modifying default version strings of DNS servers. We record DNS server fingerprints by collecting interaction behaviour of different DNS servers. After that, we propose a novel DNS server fingerprint inferring model based on fpdns-ng algorithm. We apply machine learning methods like Decision Tree, KNN and Multilayer Perceptron to classify different kinds of DNS servers. Experiment results prove the effectiveness of our method and show that our approach is more accurate than the state-of-art in fingerprinting mainstream DNS servers.

KEYWORDS

DNS Server Fingerprinting, Borderline Behaviour Analysis, Supervised Learning.

Unlocking Efficiency and Sustainability Inenzymeoptimization: a Machine Learning-drivenapproachfor Industrial Applications

Yijia Zhang1, Ivan Revilla2, 1Huaai Preparatory Academy, No. 188 Jingyuan West Road, Pidu District, Chengdu, Sichuan, China 610000, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768

ABSTRACT

This paper addresses the challenge of optimizing enzyme activity and production in various industries by leveragingmachine learning models [9]. Traditionally, enzyme optimization has been resource-intensive and costly [10]. Ourproposed solution involves collecting diverse enzymatic reaction data, generating synthetic data, and using cross- validation and ensemble methods for model selection. Challenges such as data availability and negative valuegeneration in dummy data were addressed creatively. Experimentation revealed that ensemble methods like RandomForest and Decision Tree Regressor outperformed linear models, highlighting the potential of machine learninginenzyme optimization [11]. This research of ers a data-driven approach that promises ef iciency and resourceconservation, with significant implications for biotechnologists, industrial manufacturers, and the scientificcommunity [12]. The application of machine learning in enzyme optimization not only streamlines processes but also paves the way for sustainability and innovation in enzyme-related industries, making it a compelling solutionfor widespread adoption.

KEYWORDS

Enzyme Optimization, Machine Learning Models, Resource Conservation, Industrial Ef iciency.

Towards Optimizing Performance of Machine Learning Algorithms on Unbalanced Dataset

Asitha Thumpati and Yan Zhang, School of Computer Science and Engineering, California State University San Bernardino, USA

ABSTRACT

Imbalanced data, a common occurrence in real-world datasets, presents a challenge for machine learning classification models. These models are typically designed with the assumption of balanced class distributions, leading to lower predictive performance when faced with imbalanced data. To address this issue, this paper employs data preprocessing techniques, including Synthetic Minority Oversampling Technique (SMOTE) for oversampling and random undersampling, on unbalanced datasets. Additionally, genetic programming is utilized for feature selection to enhance both performance and efficiency. In our experiment, we leverage an imbalanced bank marketing dataset sourced from the UCI Machine Learning Repository. To evaluate the effectiveness of our techniques, we implement it on four different classification algorithms: Decision Tree, Logistic Regression, K-Nearest Neighbors (KNN), and Support Vector Machines (SVM). We compare various evaluation metrics, such as accuracy, balanced accuracy, recall, F-score, Receiver Operating Characteristics (ROC) curve, and Precision-Recall (PR) curve, across different scenarios: unbalanced data, oversampled data, undersampled data, and data cleaned with Tomek-Links. Our findings reveal that all four algorithms demonstrate improved performance when the minority class is oversampled to half the size of the majority class and the majority class is undersampled to match the minority class. Subsequently, applying Tomek-Links on the balanced dataset further enhances performance.

KEYWORDS

Unbalanced Dataset, Oversampling, Undersampling, Feature Selection, Classification.

A Data-driven Strategy for Online Hate Speech Spreader Identification Using Modified Pagerank

Smita Ghosh and Shiv Jhalani, Department of Mathematics and Computer Science, Santa Clara University, Santa Clara, California, USA

ABSTRACT

Social media platforms have become breeding grounds for the dissemination of misinformation and harmful content, including hate speech. This research paper aims to tackle the urgent problem of hate speech circulation on Online Social Networks by primarily focusing on the early identification of users who are prone to spreading it. To achieve this objective, a novel data-driven metric is introduced, referred to as ’Hate Speech Potential’. Additionally, an innovative approach is proposed that leverages a modified version of the PageRank algorithm, termed the ’Hate Speech Potential Rank’ algorithm, to effectively detect and identify malicious users within a network. In a vast network with billions of nodes, the rapid spread of content makes timely detection and mitigation crucial. By assessing a user’s past behaviour of sharing or publishing hate speech, their ’Hate Speech Potential’ can be determined, enabling the identification of sources and spreaders of such content. The modified PageRank algorithm considers both the user’s individual characteristics and the influence of their neighbourhood, thereby capturing a more comprehensive picture of their sharing patterns. A pre-trained machine learning model was employed to accurately classify hate speech posts. By combining the predicted labels and user characteristics and implementing the modified PageRank algorithm, this paper aims to gain deeper insights into the dynamics of information dissemination within a social network, thereby contributing to a better understanding of user sharing behaviour and facilitating the development of effective strategies for addressing hate speech. K-Means clustering was used in experimental evaluations, demonstrating the effectiveness of the proposed approach.

KEYWORDS

Hate Speech Spreader Detection, PageRank, Social Networks, Machine Learning.

Integrating Iot and Blockchain Technology for Securing Sensors Communication

Sonali B. Wankhede1, Dhiren Patel2, Mahesh Shirole3, 1, 3VJTI, Mumbai, India, 2SVNIT, Surat, India

ABSTRACT

Blockchain offers new approaches to manage and store data. The technical trajectory of blockchain makes it beneficial in various industries because of its benefits for tamperproof and immutable data security. The increasing use of IoT devices will surely result in a smarter world. A growing number of applications, including environment monitoring, building automation, smart metering, surveillance, and asset tracking, are turning to networks of wireless sensors, each of which is capable of a combination of processing, communication, and sensing. The Internet of Things (IoT) has facilitated the transition to a connected and intelligent digital environment, allowing smart manufacturing, improved decision-making and data analytics. In this study, we review the various IoT and Blockchain frameworks and discuss how integration of IoT and Blockchain resolves existing gaps.

KEYWORDS

Blockchain, IoT, IIoT, Sensors, Blockchain-IoT framework.

Enhancing Clinical Trial Access and Diversity: an Innovative App for Personalized Trial Recommendations and Equitable Healthcare Participation

Jasmine Xing1, Ang L2, 1Buckingham Browne and Nichols Schoo, 80 Gerrys Landing Road, Cambridge, MA 02138, 1Computer Science Department, California State Polytechnic University, Pomona, CA 91768

ABSTRACT

This project focuses on creating an app to recommend clinical trials to users based on their profiles, improving access and diversity in trial participation. It addresses disparities in trial representation and aims to provide easy access to trial information. The app utilizes web scraping, a recommendation system, and a saving mechanism for trials. An experiment evaluates the recommendation systems accuracy, highlighting the challenges of ethnicity clustering. The app aims to promote equitable healthcare access and expand participation in clinical trials. Shortcomings of existing solutions include limited scope, such as being specific to cancer trials or certain patient groups. In contrast, the app aims to provide recommendations for a broader range of diseases and profiles, improving accessibility and diversity. Future improvements include refining the recommendation model, increasing the number of clusters, adding location-based recommendations, and expanding the trial database. The ultimate goal is to enhance the diversity of trial participants and improve healthcare equity.

KEYWORDS

Clinical Trial Recommendations, Healthcare, Trial Participation, Personalized Healthcare Acce.

Enhancing Transfer Learning Across Annotation Schemes With Minerr: a Novel Metric

Samuel Guilluy1, Florian Méhats2, Billal Chouli3, 1IRMAR, Université Rennes, Rennes, 2IRMAR, Ravel Technologies on leave from Univ Rennes, Rennes, 3Headmind Partners AI & Blockchain, Paris

ABSTRACT

This paper introduces MINERR (MINimal ERRor evaluation metric between consecutive tasks), a novel metric designed to enhance the efficiency of transfer learning in the context of argument structure identification. One of the principal challenges in the Argument Mining field pertains to the need for high-quality training data, which requires achieving a high level of inter-annotator agreement for argument constituents. As a consequence, datasets within this domain tend to be smaller compared to those in other domains. To address this issue, we propose the consolidation of different datasets and employ the classical two-step method for argument identification, encompassing the identification of argumentative spans and the categorization of labels. An issue related to the separation of these two tasks is the errors interconnectedness between them. To tackle this problem, we introduce a new metric that distinguishes errors stemming from incorrect labeling and errors arising from span misidentification. Our approach incorporates a novel method for dissecting the prediction error of an argument component labeling task into two distinct categories: errors caused by misidentifying the component and errors resulting from assigning incorrect labels. Subsequently, we evaluate our method using a corpus including four distinct argumentation datasets. Overall, this work facilitates the development of a new transfer learning methodology for the application of diverse argument annotation schemes.

KEYWORDS

Argument Mining · Natural Language Processing · Artificial Intelligence.

Saving Endangered Languages With a Novel Three-way Cycle Cross-lingual Zero-shot Sentence Alignment

Eugene Hwang, BlueCore Labs, The Masters School, New York, USA

ABSTRACT

Sentence classification, including sentiment analysis, hate speech detection, tagging, and urgency detection is one of the most prospective and important subjects in the Natural Language processing field. With the advent of artificial neural networks, researchers usually take advantage of models favorable for processing natural languages including RNN, LSTM and BERT. However, these models require huge amount of language corpus data to attain satisfactory accuracy. Typically this is not a big deal for researchers who are using major languages including English and Chinese because there are a myriad of other researchers and data in the Internet. However, other languages like Korean have a problem of scarcity of corpus data, and there are even more unnoticed languages in the world. One could try transfer learning for those languages but using a model trained on English corpus without any modification can be sub-optimal for other languages. This paper presents the way to align cross-lingual sentence embedding in general embedding space using additional projection layer and bilignual parallel data, which means this layer can be reused for other sentence classification tasks without further fine-tuning. To validate power of the method, further experiment was done on one of endangered languages, Jeju language. To the best of my knowledge, it is the first attempt to apply zero-shot inference on not just minor, but endangered language so far.

KEYWORDS

Natural Language Processing, Large Language Models, Transfer Learning, Cross-lingual Zero-shot, Embedding Alignment, BERT, Endangered Languages, Low-resourced Languages.