D Gaussian Splatting (3DGS) algorithms have gained increasing attention due to their ability to enable realistic 3D scene reconstruction with faster runtime. However, training these models on graphical processing units (GPU) faces unique challenges during the backpropagation stage, primarily due to memory barriers in the execution model, where every pixel within a tile computes the gradient for a single Gaussian per thread. In this paper, we propose the Backward-Striding Optimized Accelerator for 3D Gaussian Splatting (BOA-3DGS), a hardware-software co-optimized accelerator designed to optimize backward rasterization of 3D Gaussian Splatting by enabling pixels to stride through and select only relevant Gaussians for computing. However, such processing styles lead to a large number of memory stalls due to conflicting gradient storage and Gaussian parameter fetches. By introducing a pixel-independent, funnel-like multi-Gaussian alpha computation and a majoritybased gradient accumulation method, we avoid such memory stalls to efficiently accelerate backpropagation of 3DGS. Together, these enhancements improve gradient calculation efficiency and accelerate the backpropagation stage of 3DGS rasterization. Experimental results demonstrate that BOA-3DGS achieves up to $1.58 \times$ speedup during backpropagation compared to prior work, while utilizing only $0.86 \times$ the area and consuming $0.99 \times$ power.

Predicting the success of marketing campaigns is a critical challenge in the fast-moving business world. This challenge requires advanced models that can help deal with complicated consumer behavior and tell us what actions to take. The goal of this study is to create and test a hybrid deep learning model that predicts campaign success using feature embeddings and dense neural networks. The hybrid model uses TabNet in its analysis to identify key features. The analysis identifies NumWebPurchases, MntGoldProds, Teenhome, Income, and Recency as some of the key predictors. Moreover, through which this knowledge is integrated into its learning frameworks. The hybrid model performed better with $91.24 \%$ accuracy, $89.76 \%$ precision, $89.32 \%$ recall, $89.54 \%$ F1-score compared to Multi-Layer Perceptron (MLP), TabNet and Deep Belief Networks (DBN). Presently, the model’s capability to reduce false negatives assists target differentiations in ways that present-day methods do not. Highlighting the relevance of examining feature importance, this exploration also gives marketers a chance to grasp consumer behavior and market trends which could be seen as beneficial.

This study employs network graph analysis to explore the social interactions and relationships between characters in Jane Austen’s Pride and Prejudice. By constructing a network graph where each node represents a character and edges denote the frequency and nature of their interactions, the analysis provides a quantitative and visual representation of the novel’s social structure. The resulting network graph reveals the centrality and influence of various characters, highlighting key relationships and social dynamics within the narrative. Through this approach, the study not only enhances our understanding of character relationships but also demonstrates the utility of network analysis in literary studies, offering new insights into the interpersonal complexities of Pride and Prejudice.

Fraud or Fake financial transactions seriously impact digital payment systems, necessitating more advanced detection mechanisms to mitigate the associated risks. Fraud trends that are always changing have made the traditional methods used to identify fraud cases obsolete, such as rule-based fraud detection and machine learning models. Recent studies have shown that Graph Neural Networks (GNNs) can better capture the relationship between financial transactions, while transformers are effective at recognizing sequential fraud patterns. Yet, the existing models that incorporate both do not perform well in this manner. To fill this gap in existing research, we have created a new model for detecting fraudulent transactions, the Hybrid GNN-Transformer Fraud Detection Model. This uses graphbased learning along with deep sequential feature extraction to better distinguish frauds from genuine transactions. The hybrid model had better performance compared to single models such as autoencoders, GNNs, and LSTMs, getting an accuracy of $99 \%$, as well as a precision of.99 and a recall of 1.00 when it comes to detecting fraudulent transactions. Comparisons show that GNNs and LSTMs still, when combined with transformers, there is an improved ability in the identification ofare able to capture key transaction interdependencies on their own. Still, when combined with transformers, they have an improved ability to identify complicated fraud activities.

Air pollution is still a major problem in cities where pollutants, such as ozone (O3) and sulfur dioxide (SO2), can harm health and the environment. Thus, being able to forecast the Air Quality Index (AQI) can help make better-informed decisions and interventions. This study assesses traditional machine learning (ML), deep learning (DL), and hybrid models in AQI prediction using real-world data from New York City from 2014 to 2015. The research involved comparing various machine learning models such as Random Forest, XGBoost, and Support Vector Regression, as well as Long Short-Term Memory (LSTM) networks and hybrid models combining ML with DL techniques. Model performance was evaluated based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination ($\mathbf{R}^{2}$) values in multiple neighborhoods. The authors found that Random Forest and XGBoost performed better than stand-alone LSTM on both accuracy of forecasting and stability. A hybrid model with Random Forest and LSTM was also more robust than the stand-alone LSTM models in all neighborhoods.

Digital Content protection is important Ensuring document integrity in the digital age is a critical challenge. Traditional visible and invisible watermarking methods offer partial protection but have limitations. Visible watermarking can be removed and degrade quality, while invisible watermarking requires specialized tools and may degrade through compression. TThis study presents an innovative document authentication framework that leverages part-of-speech (POS) tag-based graph representations. By transforming textual data into structured graphs, the method captures the unique syntactic signature of each document. This linguistic fingerprint is then encrypted and stored independently, enabling robust verification of document authenticity. The approach not only reinforces security and integrity but also seamlessly integrates with existing natural language processing (NLP) pipelines.

Credit card fraud detection is a critical challenge in modern financial systems, requiring robust and efficient solutions to identify suspicious transactions accurately. This study addresses the issue by utilizing machine learning techniques to detect potentially fraudulent transactions. A key focus of the research is the handling of imbalanced datasets, where genuine transactions vastly outnumber fraudulent ones. To address this imbalance, we increased the sample size of fraudulent data using oversampling techniques and subsequently applied four distinct machine learning models to assess their performance. Through iterative experimentation, we identified the optimal magnification ratio that enhances the model’s ability to distinguish between legitimate and fraudulent transactions. The results demonstrate that balancing the dataset significantly improves detection accuracy, providing insights into effective model configurations for realworld applications. This research contributes to the development of more reliable and efficient fraud detection systems in the financial sector.

This study explores quantum computing and blockchain, focusing on quantum-safe algorithms. As quantum computing progresses, it threatens blockchain cryptography, necessitating quantum-resistant/safe algorithms. Using statistical analysis and simulations to address future cyber threats, we analyze quantum-safe cryptography in blockchains. We pioneer machine learning models to assess quantum-safe algorithm performance across encryption, key generation, signatures, speed, scalability, energy efficiency, and security. Our analysis, referencing Ethereum’s Ether values, pinpoints areas for quantum-safe cryptography improvements, addressing limitations and proposing solutions to strengthen security and efficiency. In experiments, we observed a 15% increase in encryption strength, 10% enhanced key generation efficiency, and improved reliability parameters. These findings stress the importance of machine learning in quantum-safe cryptography for widespread adoption and long-term security against quantum threats. In conclusion, our research paves the way for the integration of quantum-safe algorithms, ensuring the resilience of blockchain systems in the face of quantum advancements.

Accurate and automated sound classification enables a strong groundwork for diverse advanced deep learning applications within the audio and music domain. This study focuses on the application of Convolutional Neural Networks (CNN) and combined LSTM (Long Short-Term Memory) and GRU (Gated Recurrent unit) models for instrument classification from audio signals, contributing to intelligent audio processing systems. Our proposed model exclusively utilizes the Mel-frequency cepstral coefficients (MFCCs) extraction from the audio data for preprocessing. A large and complex dataset, including Nineteen instrument classes are used for training and evaluation. These experimental results demonstrate promising performance, with our proposed CNN architecture achieving an impressive accuracy of 97%, and the LSTM-GRU model achieves a lower accuracy of 80%, compared to the CNN model on the multi-label sound classification task for instruments classes, but its ability to model temporal dependencies add valuable insights into the dynamics of instrument audio sequences. These findings provide valuable insights for researchers and practitioners in audio signal processing and machine learning.

In the realm of Decentralized Finance (DeFi), this manuscript introduces a Hybrid Cross-Chain Model. As DeFi architectures grapple with the complexities of monolithic single-chain platforms, our proposed model orchestrates a symphony of multiple chains to facilitate seamless cross-chain communication, offering a poised solution to scalability and transaction speed challenges. Incorporating modeling effects and simulations, our rigorous performance evaluation underscores the model's excellence and includes an in-depth analysis of its performance, particularly focusing on robust security measures. The model is positioned as a cornerstone in an interconnected DeFi landscape by emphasizing stringent measures to ensure data integrity and uphold consensus mechanisms. User-centric enhancements promise swift transaction confirmations and reduced fees, improving the overall experience. The abstract culminates with a comparative analysis, positioning the Hybrid Cross-Chain Model as an innovative solution with profound implications for the future of DeFi. This manuscript advocates for ongoing research and development, heralding a new era of sophistication and resilience in decentralized finance.

This research presents a sophisticated technological framework for Cross-Chain Decentralized Finance (DeFi) and Smart Contract systems by seamlessly integrating Markov Models, Brownian Motion, and Stationary Processes. Focused on enhancing the adaptability and efficiency of financial interactions across interconnected blockchain networks, this framework establishes the foundational elements necessary for dynamic system modeling. The incorporation of Markov Models captures state transitions, Brownian Motion models random fluctuations, and Stationary Processes ensure statistical stability. The paper explores the technological implications of these stochastic processes, addressing challenges in system interoperability, latency, and security within decentralized financial ecosystems. Envisioning a future where decentralized systems are optimized and resilient, the research investigates advancements in blockchain protocol design, consensus mechanisms, and transaction validation strategies. The proposed framework, influenced by the dynamic and statistical nature of Brownian Motion and Stationary Processes, underscores the need for robust data structures, real-time data feeds, and decentralized oracle networks. This research invites collaboration from the blockchain, smart contract, and stochastic modeling communities to contribute to the ongoing exploration and refinement of this powerful technological framework, poised to reshape the landscape of cross-chain financial technologies.

This study assesses the capacity of artificial intelligence (AI) algorithms to mimic a human's sense of taste, specifically in the context of wines. We utilize wine data and machine learning tests to compare the performance of ML models, including Decision Tree, Random Forest, Logistic Regression, and Support Vector Machine, against that of a real sommelier. Our findings show that the Random Forest model outperforms all others in accuracy. Moreover, our results uncover new insights into wine tasting. While our human tongue can detect 11 variables from our dataset, only four of these variables are used by the brain to discern wine flavors. This discovery challenges the previously held belief in the complexity of our sensory system. Our methodology paves the way for future research by streamlining data collection and enhancing its cost efficiency and accuracy by focusing on these essential variables rather than the entire set of eleven.

Intrusion Detection System (IDS) is a crucial security mechanism for protecting computer networks from cyber-attacks. Deep learning models have the potential to detect attack types by leveraging their ability to learn and extract features from large volumes of data. In this study, we compare the performance of four different deep learning algorithms for IDS: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), bidirectional LSTM, and bidirectional GRU. We evaluate the attack prediction accuracy for three types of attacks: Denial of Service (DoS), Generic, and Exploits. We vary each algorithm's range parameter and epochs and determine the best parameter combination sets for achieving the highest accuracy. Our experimental results demonstrate that increased range parameters influence the accuracy of LSTM, bi-LSTM, and Bi-GRU models. Ultimately, GRU proved to have the most outstanding performance among the four algorithms tested.

This study utilizes vibration-based signal analysis as a non-destructive testing technique that involves analyzing the vibration signals produced by a structure to detect possible defects or damage. The study aims to employ deep learning models to identify defects in a 3D-printed cantilever beam by analyzing the beam’s tip displacement given a random input signal generated by an electromagnetic shaker. This study is focused on the output signal without any information of the random input, which is common for structural health monitoring applications in practice. Additionally, the study has revealed that the number of times the test set is applied to the trained model significantly impacts the accuracy of the model’s consistent predictions.

Applying artificial intelligence and machine learning to analyzing network traffic has the potential to be transformative in protecting organizations from cyber threats. Intrusion detection systems (IDS) are historically rule-based; however, they could be improved. Applying machine learning in the form of Anomaly Detection could be the next step in preventing cyber threats from causing malicious activity on the network. Two algorithms that are implemented in anomaly detection through the use of Amazon SageMaker are Random Cut Forest (RCF) and XGBoost. The data for this project are the training and testing data set provided by the UNSW-15 data set. The models are created using the Jupiter Notebook on the Amazon SageMaker Studio Lab platform. The models were tested using the metrics of accuracy, precision, recall, and F1 score. The best-performing model was the XGBoost model, with an accuracy of 61.83%. The recall for this model was 96.49%, and the f1 score was 73.24%.

Containerization is becoming an increasingly common aspect of DevOps. Adding a container layer increases the complexity and could impact system performance. This study explores the performance differences of the Apache and Nginx web servers on Virtual Machines (VMs) and Docker Containers with official web server images from Docker Hub. A sandbox environment was created with both containerized and non-containerized versions of the web servers, and their performance was analyzed using line graphs. The results showed differences in performance between VMs and Docker Containers, with some variation from previous research due to the virtualization being done locally rather than on the cloud. This study would be advantageous for organizations with on-premises infrastructure due to security or governing regulations.

Machine learning classifiers are known algorithms used to classify network intrusion detection due to the drastic growth of data, new tools are being required to handle such a large amount of data within a short time frame. In this Paper, we present a Model using the Apache Mahout Framework to train machine learning classifiers Random Forest (RF), Logistic Regression (LR), and Naive Bayes (NB) on CSE-CIC-IDS2018 dataset using Chi-Square and ANOVA f-test filter-based feature selection technique on an Apache Hadoop Framework. The performance of classifiers is measured in terms of Accuracy, Kappa, Precision, Recall, and Fl-Score for a comparative analysis of the various machine learning classifiers.

This study proposes a visualization framework for bottlenose dolphin (Tursiops truncatus) whistle classification using statistical signal properties such as slope, standard deviation, kurtosis, skewness, minimum and maximum frequency, frequency range, absolute value, and duration. For each whistle, all statistical properties are stored in a MySQL database, one of the relational databases, so that users can classify whistles. Moreover, correlation values are provided in a heat map to examine the similarity of whistles among the population. This framework will allow marine mammal researchers to determine the whistle properties of individual bottlenose dolphins. Researchers can then investigate behavioral and ecological questions such as (i) does whistle structure differs among dolphin populations (ii) and does anthropogenic noise affects whistle structure.

As many artificial neural libraries are developing the deep learning algorithm and implementing it became accessible to anyone. This study points out the disparity of performance in deep learning models such as convolutional neural networks (CNN) when implemented with different artificial neural libraries. Libraries such as Keras, Pytorch, and MXnet was utilized for each three CNN model then binary image classification was done based on the Dogs vs. Cats dataset from Kaggle. With using 75% of the dataset as the training set and the rest of 25% as a testing set, and as a result, each CNN model gave a different F1 score value and accuracy.

Structural health monitoring provides significant and obvious potential in enhancing our life safety and extending the service life of systems. In practice, structures are often exposed to random excitations without knowing the exact characteristics of its source. This paper proposes a novel method of the implementation of LSTM and GRU models for characterizing anomalies in output-only random vibration signals. In the proposed approach, the time response of a 3D printed PLA beam is measured when subjected to a random excitation and used to train LSTM and GRU models. Healthy time response and four additional cases that contain a small mass at varied locations along the beam are used as model inputs. These inputs represent normal and abnormal signals which are then classified to diagnose the health state of the structure. In this study, the random vibration time responses with large amplitude (so-called signal caricature) were selectively employed for creating the output-only models. The results illustrate the signal caricature data set leading to both more accurate and efficient characterization of the structural health state, compared to utilizing the entire time response as an input to the models. Modal properties obtained by traditional vibration analysis support the effectiveness of utilizing the signal caricature with LSTM/GRU models, demonstrating great potential in identifying defects in practical applications.

Structural health monitoring is an area of growing interest and is worthy of new and innovative approaches. Since the automatic diagnosis of structures is very complex and challenging, recent research to apply deep learning techniques has been actively conducted. In this study, we assumed that a PLA beam copied by 3D printing is the smallest unit constituting a complex structure and applied GRU to detect defects. To set the defect point of the beam, a total of four holes were drilled at regular intervals, and then a mass was attached. Signals at different locations were collected through a vibrator and trained through GRU, and the results were compared in terms of RMSE value. As a result of this experiment, we checked the defect by inputting test data into the trained model and were able to measure the defect degree of the PLA beam with a weighted average F1 score of 84%.

Vibration-based methods have received considerable attention in structural condition monitoring applications. We have proposed a model to detect damaged points of a target structure using the GRU model and classify the 0.84 overall accuracy. To increase the model's accuracy in this research, we propose an ensemble deep learning model using LSTM and bi-directional LSTM incorporated with GRU. Each model predicted its RMSE trend and combined the damage estimation results from both models, which are mostly close to the true damage locations. As a result of synthesizing the three algorithms, the damage point of the cantilever beam was found with an accuracy of 0.88 and a misclassification rate of 0.12. The results indicate that the proposed combined approach provides enhanced reliability than a single algorithm.

This research proposes the methodology to extract common configuration parameter set by applying multiple benchmark applications including TeraSort., TestDFSIO, and MrBench on the Hadoop Distributed File System. In the process of determining parameter set for each stage, one parameter and its associated values selected which is reduced system performance in terms of overall execution time difference are measured by multiple applications on a Hadoop cluster. The experimental results demonstrate the proposed extended greedy manner provide a feasible benchmark model for the multiple tasks. In this way, we have found several parameter value sets that can reduce the execution time by 27% of the values provided by Hadoop default.

As various wireless devices share the same frequencies in the unlicensed 2.4-GHz industrial scientific medical band, frequency sharing has become a challenging issue in the heterogeneous network. Many Wi-Fi applications increase network traffic and lead to the significant performance loss of other protocol devices including ZigBee for critical missions (e.g., medical devices) in the same band. In this paper, we propose a coexistence solution of the guide busy tone (GBT), providing reliable communications to the ZigBee network under Wi-Fi interference, and present fairness criteria in the tradeoff relation between Wi-Fi and ZigBee with GBT. The proposed GBT design, consisting of a GBT signaler and a busy tone canceller, reserves a channel for ZigBee through the full-duplex technique under heavy Wi-Fi traffic. Our experimental evaluation shows that the packet delivery ratio of the ZigBee network can be improved up to nearly 100% under the saturated Wi-Fi traffic by using GBT, which is scalable for the multinode case as well.

Security has become an increasingly important issue in wireless communications for the IoT (Internet of Things) environments, to which physical layer approaches can contribute by differentiating desired transceiver and wiretap channels for security of confidential data. In this paper, we propose an optimal power allocation strategy for practical physical-layer security, based on friendly jamming with cancellation for anti-eavesdropping. In particular, secrecy outage probability is evaluated in scenarios involving a pair of transmitter-receiver and a passive eavesdropper near the receiver. We derive the optimal power allocation strategy and the required cancellation capability to enhance secrecy performance. Numerical results verify the tradeoff between jamming power ratio and cancellation capability. Furthermore, our proposed scheme can achieve the improved secrecy rate regardless of availability of eavesdropper channel information.

This paper proposes a method to facilitate the identification process for a set of configuration parameters to achieve the optimal performance with respect to a benchmark program in HDFS in an automated manner. Performance optimization of Hadoop processes is a tedious yet challenging problem due to the complexity of the systems organization with an extensive list of configuration parameters to be considered. An Automated Benchmarking Configuration Method (ABCM) is developed in this work to facilitate the identification process for the set of configuration parameters that minimizes the execution time of a benchmark, namely TestDFSIO Write and Read in particular. A two-phased configuration parameters selection process with a simple sampling technique is proposed in order to mediate the exponential computation time otherwise. By using the proposed technique, we have automatically found the sets of top five selected optimal configuration parameters that reduced the average execution time by 32% compared to the execution time with the default set of Hadoop configuration parameters.

Current implementation of Hadoop is based on an assumption that all the nodes in a Hadoop cluster are homogenous. Data in a Hadoop cluster is split into blocks and are replicated based on the replication factor. Service time for jobs that accesses data stored in Hadoop considerably increases when the number of jobs is greater than the number of copies of data and when the nodes in Hadoop cluster differ much in their processing capabilities. This paper addresses dynamic data rebalancing in a heterogeneous Hadoop cluster. Data rebalancing is done by replicating data dynamically with minimum data movement cost based on the number of incoming parallel mapreduce jobs. Our experiments indicate that as a result of dynamic data rebalancing service time of mapreduce jobs were reduced by over 30% and resource utilization is increased by over 50% when compared against Hadoop.