ITNG 2023 20th International Conference on Information Technology-New Generations
- 2023
- Book
- Editor
- Shahram Latifi
- Book Series
- Advances in Intelligent Systems and Computing
- Publisher
- Springer International Publishing
About this book
This volume represents the 20th International Conference on Information Technology - New Generations (ITNG), 2023. ITNG is an annual event focusing on state of the art technologies pertaining to digital information and communications. The applications of advanced information technology to such domains as astronomy, biology, education, geosciences, security, and health care are the among topics of relevance to ITNG. Visionary ideas, theoretical and experimental results, as well as prototypes, designs, and tools that help the information readily flow to the user are of special interest. Machine Learning, Robotics, High Performance Computing, and Innovative Methods of Computing are examples of related topics. The conference features keynote speakers, a best student award, poster award, service award, a technical open panel, and workshops/exhibits from industry, government and academia. This publication is unique as it captures modern trends in IT with a balance of theoretical and experimental work. Most other work focus either on theoretical or experimental, but not both. Accordingly, we do not know of any competitive literature.
Advertisement
Table of Contents
-
Frontmatter
-
Machine Learning
-
Frontmatter
-
Chapter 1. Loop Closure Detection in Visual SLAM Based on Convolutional Neural Network
Fabiana Naomi Iegawa, Wagner Tanaka Botelho, Tamires dos Santos, Edson Pinheiro Pimentel, Flavio Shigeo YamamotoThe chapter explores the critical role of loop closure detection in Visual SLAM for autonomous navigation in mobile robotics. It delves into the challenges of false positives and false negatives, and how Deep Learning, specifically CNN, can address these issues. The proposed system uses transfer learning to train a VGG-16 network on virtual images, improving feature extraction and descriptor generation. The system is validated in a Gazebo simulation, demonstrating high accuracy in loop closure detection. The chapter also highlights the importance of efficient image-to-image matching and the use of a 512-channel tensor for descriptors, reducing processing time and computational overhead.AI Generated
This summary of the content was generated with the help of AI.
AbstractIn Robotics, autonomous navigation has been addressed in recent years due to the potential of applications in different areas, such as industrial, comercial, health and entertainment. The capacity to navigate, whether autonomous vehicles or service robots, is related to the problem of Simultaneous Localization And Mapping (SLAM). Loop closure, in the context of Visual SLAM, uses information from the images to identify previously visited environments, which allows for correcting and updating the map and the robot’s localization. This paper presents a system that identifies loop closure and uses a Convolutional Neural Network (CNN) trained in Gazebo simulated environment. Based on the concept of transfer learning, the CNN of VGG-16 architecture is retrained with images from a scenario in Gazebo to enhance the accuracy of feature extraction. This approach allows for the reduction of the descriptors’ dimension. The features from the images are captured in real-time by the robot’s camera, and its control is performed by the Robot Operating System (ROS). Furthermore, loop closure is addressed from image preprocessing and its division in the right and left regions to generate the descriptors. Distance thresholds and sequences are defined to enhance performance during image-to-image matching. A virtual office designed in Gazebo was used to evaluate the proposed system. In this scenario, loop closures were identified while the robot navigated through the environment. Therefore, the results showed good accuracy and a few false negative cases. -
Chapter 2. Getting Local and Personal: Toward Building a Predictive Model for COVID in Three United States Cities
April Edwards, Leigh Metcalf, William A. Casey, Shirshendu Chatterjee, Heeralal Janwa, Ernest BattifaranoThe chapter delves into the correlation between Twitter activity and COVID-19 cases in Miami, Las Vegas, and Seattle, aiming to predict outbreaks through social media data. It compares tweet volumes with case numbers, revealing that individual user responses (replies) are more predictive than retweets or mainstream news. The study uses statistical methods and machine learning models, such as linear and polynomial regression, to forecast case counts up to a month in advance. The findings suggest that social media can be a powerful tool for real-time surveillance and understanding the spread of information during a global pandemic.AI Generated
This summary of the content was generated with the help of AI.
AbstractThe COVID-19 pandemic was lived in real-time on social media. In the current project, we use machine learning to explore the relationship between COVID-19 cases and social media activity on Twitter. We were particularly interested in determining if Twitter activity can be used to predict COVID-19 surges. We also were interested in exploring features of social media, such as replies, to determine their promise for understanding the views of individual users. With the prevalence of mis/disinformation on social media, it is critical to develop a deeper and richer understanding of the relationship between social media and real-world events in order to detect and prevent future influence operations. In the current work, we explore the relationship between COVID-19 cases and social media activity (on Twitter) in three major United States cities with different geographical and political landscapes. We find that Twitter activity resulted in statistically significant correlations using the Granger causality test, with a lag of one week in all three cities. Similarly, the use of replies, which appear more likely to be generated by individual users, not bots or public relations operations, was also strongly correlated with the number of COVID-19 cases using the Granger causality test. Furthermore, we were able to build promising predictive models for the number of future COVID-19 cases using correlation data to select features for input to our models. In contrast, significant correlations were not identified when comparing the number of COVID-19 cases with mainstream media sources or with a sample of all US COVID-related tweets. We conclude that, even for an international event such as COVID-19, social media tracks closely with local conditions. We also suggest that replies can be a valuable feature within a machine learning task that is attempting to gauge the reactions of individual users. -
Chapter 3. Integrating LSTM and EEMD Methods to Improve Significant Wave Height Prediction
Ashkan Reisi-Dehkordi, Alireza Tavakkoli, Frederick C. Harris JrThe chapter delves into the integration of Long Short-Term Memory (LSTM) networks and Ensemble Empirical Mode Decomposition (EEMD) for improving Significant Wave Height (SWH) prediction. It begins by highlighting the importance of SWH in harnessing ocean wave energy, a renewable resource with high energy density. The authors review existing methods for predicting oceanographic parameters and wave heights, noting the limitations of statistical, numerical, and empirical models. They introduce LSTM networks as a powerful tool for time series prediction, particularly in handling non-linear data. The methodology section describes the development of two LSTM models: one using a sequence of wave heights and another incorporating EEMD for time-frequency data analysis. The EEMD-LSTM model decomposes wave height time series into intrinsic mode functions, which are independently learned by the LSTM. The results show that the EEMD-LSTM framework significantly outperforms the non-EEMD LSTM model, with improvements in various error indices across different forecasting windows. The chapter concludes by emphasizing the benefits of wavelet decomposition and reconstruction by EEMD, and suggests future work involving the integration of numerical simulation methods for enhanced robustness.AI Generated
This summary of the content was generated with the help of AI.
AbstractOne of the most significant reliable and renewable energy sources is wave energy which has the most energy density among the renewable energy sources. Significant Wave Height (SWH) plays a major role in wave energy and hence this study aims to predict wave height using time series of wave characteristics as input to various machine learning approaches and analyze these approaches under several scenarios. Two different machine learning algorithms will be implemented to forecast SWH. In the first approach, the SWH will be forecasted directly using a Long Short Term Memory (LSTM) network and in the second approach an LSTM and an Ensemble Empirical Mode Decomposition (EEMD) method are proposed for SWH prediction. For this purpose, the elements of wave height will be initially decomposed and used for training an LSTM network to calculate the time series of SWH. Also, the calibration and verification of the modeled wave characteristics will be done using real data acquired from buoys. The results imply that the EEMD approach provides more accurate results and calculating the wave height through the decomposition and prediction of its main wave components can deliver more accurate outcomes considering various error indices. Also, it can be inferred from the results that the accuracy of the predictions will decrease as the forecasting time horizon increases. -
Chapter 4. A Deep Learning Approach for Sentiment and Emotional Analysis of Lebanese Arabizi Twitter Data
Maria Raïdy, Haidar HarmananiThe chapter delves into the application of deep learning techniques to analyze sentiment and emotions expressed in Lebanese Arabizi tweets during pivotal events such as the 2019 social unrest and the 2020 economic crisis. It outlines the creation and curation of a unique dataset of Lebanese Arabizi tweets, highlighting the challenges and methods used in text preprocessing and feature extraction. The study compares the performance of various machine learning models with a deep learning approach, showcasing the effectiveness of the latter in sentiment and emotion classification. The results are then used to measure the Lebanese Social Happiness Index (LSHI), providing insights into the emotional state of the population during these significant events. The chapter concludes with a discussion on the implications of these findings and the potential applications of the proposed methodology in understanding societal well-being through social media data.AI Generated
This summary of the content was generated with the help of AI.
AbstractArabizi is an Arabic dialect that is represented in Latin transliteration and is commonly used in social media and other informal settings. This work addresses the problem of Arabizi text identification and emotional analysis based on Lebanese dialect. The work starts with the extraction and construction of a dataset and uses two machine learning models. The first is based on fastText for learning the embeddings while the second uses a combination of recurrent and dense deep learning models. The proposed approaches were attempted on the Arabizi dataset that we extracted and curated from Twitter. We attempted our results with six classical machine learning approaches using separate sentiment and emotion analysis. We achieved the highest result in literature for the binary sentiment analysis with an F1 score of 81%. We also present baseline results for the 3-class sentiment classification of Arabizi tweets with an F1 score of 64%, and for emotion classification of Arabizi tweets with an f1 score of 61%. -
Chapter 5. A Two-Step Approach to Boost Neural Network Generalizability in Predicting Defective Software
Alexandre Nascimento, Vinicius Veloso de Melo, Marcio Basgalupp, Luis Alberto Viera DiasThe chapter discusses the economic impact of software quality issues and the limitations of existing machine learning models in predicting software defects. It introduces a novel two-step learning approach using a quasi-Newton method to tune neural networks, significantly improving their performance and generalizability. The approach is validated using a comprehensive dataset from NASA projects, demonstrating its effectiveness in handling imbalanced datasets and enhancing software testing efforts.AI Generated
This summary of the content was generated with the help of AI.
AbstractWith society’s digitalization, the ever-growing dependence on software increased the negative impact of poor software quality. That impact was estimated at $2.41 trillion to the US economy in 2022. In searching for better tools for supporting quality assurance efforts, such as software testing, many studies have demonstrated the use of Machine Learning (ML) classifiers to predict defective software modules. They could be used as tools to focus test efforts on the potentially defective modules, enhancing the results achieved with limited resources. However, the practical applicability of many of those studies is arguable because of (1) the misuse of their training datasets; (2) the improper metrics used to measure those classifiers’ performance; (3) the use of data from only a system or project; and (4) the use of data from only a computer programing language. When those factors are not considered, the experiments’ results are biased towards a very high accuracy, leading to improper conclusions related to the generalizability of classifiers to practical uses. This study sheds light on those issues and points out promising results by proposing and testing the cross-project and cross-language generalizability of a novel 2-step approach for artificial neural networks (ANN) using a large dataset of 17,147 software modules from 12 projects with distinct programming languages (C, C++, and Java). The results demonstrated that the proposed approach could deal with an imbalanced dataset and outperform a similar ANN trained with the conventional approach. Moreover, the proposed approach was able to improve by 277% the number of defective modules found with the same software test effort. -
Chapter 6. A Principal Component Analysis-Based Scoring Mechanism to Quantify Crime Hot Spots in a City
Yu Wu, Natarajan MeghanathanThe chapter introduces a Principal Component Analysis (PCA)-based scoring mechanism to identify and quantify crime hot spots in a city. By analyzing historical crime data, the method provides a reliable and interpretable way to assess the severity of crime occurrences. This approach offers law enforcement agencies a flexible tool to rank areas for policing tasks, moving beyond the binary classification of crime hot spots. The study also contributes to the spatio-temporal analysis of crime, highlighting the significance of spatial features over temporal ones. The experimental results, showcased through a heat map of Boston, reveal clustered crime hot spots and their temporal stability. Despite some limitations, the research paves the way for future enhancements, including the integration of more diverse crime data and advanced data mining techniques.AI Generated
This summary of the content was generated with the help of AI.
AbstractHot spots policing is a tactic that judiciously distributes police resources in accordance with regional historical data on criminal occurrences and local crime patterns. Unquestionably, the key to this method is identifying crime hot spots. A growing number of studies are looking into how to pinpoint crime hot spots with greater accuracy. Nevertheless, the majority of them merely take the task as a binary classification problem. Our research proposes the notion of a Crime Hot Spot Score, a Principal Component Analysis (PCA)-based linear scoring mechanism for assessing regional crime severity, which equips data users with a more flexible way of utilizing crime hot spot analysis results. We conducted our study on the 3-year period crime dataset from the Boston Police Department. All our preliminary results are encouraging: we not only provide a new perspective in hot spot detection, but also reveal the correlation between the crime hot spot and its adjacent area. -
Chapter 7. Tuning Neural Networks for Superior Accuracy on Resource-Constrained Edge Microcontrollers
Alexandre M. Nascimento, Vinícius V. de Melo, Márcio P. BasgaluppThe chapter 'Tuning Neural Networks for Superior Accuracy on Resource-Constrained Edge Microcontrollers' delves into the optimization of artificial neural networks (ANNs) for deployment on low-cost, 8-bit microcontrollers. Traditional methods of increasing ANN capacity often require powerful processors, limiting their application on less capable devices. The research addresses this gap by proposing a combined optimization of weights and activation functions to enhance performance on microcontrollers. The study presents a systematic literature review and a novel methodology involving two optimization stages: training and tuning. Empirical analysis using four classification datasets demonstrates that the proposed method can significantly improve ANN accuracy on resource-constrained devices. The results show that tuned ANNs not only outperform non-tuned networks but also achieve superior performance compared to more complex networks run on computers. This chapter offers valuable insights for professionals seeking to deploy efficient and accurate ANNs on cost-effective microcontrollers, contributing to the democratization of advanced AI applications in resource-constrained environments.AI Generated
This summary of the content was generated with the help of AI.
AbstractThe approaches to tune Artificial neural networks (ANN) for running on edge devices, such as weight quantization, knowledge distillation, weight low-rank approximation, and network pruning, usually reduce their accuracy (gap 1). Moreover, they usually require at least 32-bit microcontrollers, leaving out of the equation widely used and much cheaper platforms mostly based on 8-bit microcontrollers (e.g., ATMega328p and ATMega2560), such as Arduino (gap 2). Those microcontrollers can cost between $0.01 to $0.10 on a large scale and can make viable extending IoT applications to a wider range of cheaper personal objects, such as bottles, cans, and cups. In this context, the present study addresses those two identified gaps by proposing and evaluating a technique for tuning ANN to run on 8-bit microcontrollers. 16,000 ANN with distinct configurations were trained and tuned with four widely used datasets and evaluated on two 8-bit microcontrollers. Using less than 3.5Kbytes, the embedded ANN average accuracies outperformed their benchmarks on a 64-bit computer. -
Chapter 8. A Deep Learning Approach for the Intersection Congestion Prediction Problem
Marie Claire Melhem, Haidar HarmananiThe chapter delves into the critical issue of traffic congestion at intersections, highlighting its impact on the environment and commuter welfare. It explores the evolution of smart cities and the role of IoT devices in monitoring traffic states. The authors present a deep learning approach for predicting intersection congestion, comparing the performance of Random Forests, Linear Regression, Deep Neural Networks, and LSTM models using a dataset from Kaggle and Geotab. The chapter concludes with a comparative study of these models, emphasizing the promising results of LSTM neural networks. The detailed analysis and practical applications make this chapter a valuable resource for professionals seeking to optimize traffic management in urban environments.AI Generated
This summary of the content was generated with the help of AI.
AbstractTraffic prediction at intersections is an important problem as it serves an essential role in minimizing wait time in large cities while reducing emissions. The problem is challenging, especially with spatial and temporal dependencies between intersections in a large metropolitan city. In this paper, we use a deep learning model to predict traffic congestion based on day, time and weather data. we evaluate our model using datasets from large cities including Atlanta, Philadelphia, Boston and Chicago. -
Chapter 9. A Detection Method for Stained Asbestos Based on Dyadic Wavelet Packet Transform and a Locally Adaptive Method of Edge Extraction
Hikaru Tomita, Teruya MinamotoThe chapter introduces a groundbreaking technique for detecting stained asbestos in microscopic images, addressing the challenges posed by conventional analysis methods. By utilizing Dyadic Wavelet Packet Transform (2D-DYWPT) and locally adaptive edge extraction, the proposed method extracts precise edge information from asbestos fibers, enabling accurate and efficient detection. The study compares the performance of the proposed method with fine-tuned ResNet, demonstrating superior accuracy and reliability in classifying stained asbestos. The method is validated through extensive experiments, showcasing its potential to revolutionize asbestos detection in the construction industry.AI Generated
This summary of the content was generated with the help of AI.
AbstractRecently, the development of two dye staining methods has made it easier to visually recognize asbestos. We propose a method for detecting stained asbestos-specific fiber shapes in microscopic images by extracting edge information using the two-dimensional dyadic wavelet packet transform(2D-DYWPT), which can extract detailed edge information, and the idea of eigenvalue analysis of the Hessian matrix, which captures the difference in pixel values in a locally adaptive method. At first, we convert the original image from RGB space to YIQ color space, and then apply the 2D-DYWPT to the Y and Q components. We extract 36 features depending on the statistics obtained by the 2D-DYWPT and the eigenvalue of Hessian matrices, and classify microscopic images by support vector machine. Experimental results show a comparison with fine-tuned ResNet and the results of applying the detection system to actual microscopic images. We confirmed that the performance of our method is superior to the one of ResNet in total. -
Chapter 10. Machine Learning: Fake Product Prediction System
Okey IgbonagwamThis chapter delves into the growing problem of fake product reviews on e-commerce platforms, highlighting their impact on consumer decisions and business integrity. It introduces a machine learning model using Python and the Random Forest Classifier to detect these fraudulent reviews, demonstrating higher accuracy and efficiency compared to existing methods. The chapter covers data cleaning, exploratory analysis, feature engineering, and model evaluation, providing a comprehensive approach to enhancing the reliability of online product reviews.AI Generated
This summary of the content was generated with the help of AI.
AbstractProduct review plays a vital role in hopping, especially for the online customers. Some base their buying decisions on the review; hence a fake review is a major concern (Jadhav and Gore, Int J Comput Sci Inform Technol 5(2):1447–1450, 2014). Competitions appear to facilitate Fake product malicious agents, a major challenge in the e-commerce industry. This paper intends to use machine learning to explore a predict fake or genuine products system by feeding the products into the model.
-
- Title
- ITNG 2023 20th International Conference on Information Technology-New Generations
- Editor
-
Shahram Latifi
- Copyright Year
- 2023
- Publisher
- Springer International Publishing
- Electronic ISBN
- 978-3-031-28332-1
- Print ISBN
- 978-3-031-28331-4
- DOI
- https://doi.org/10.1007/978-3-031-28332-1
Accessibility information for this book is coming soon. We're working to make it available as quickly as possible. Thank you for your patience.
