Reliable real-time vehicle control is essential for intelligent transport systems where accurate perception and decision-making depend on fast sensor data processing. This study developed a hybrid edge-cloud method integrating deep learning with Internet of Things (IoT) sensor fusion for adaptive vehicle control. Ultrasonic range data were combined with convolutional neural networks (CNNs) to enable object detection, stopping-time prediction, and braking control under varying environmental conditions. The CNN-based model was trained and evaluated under normal and simulated adverse driving scenarios. Results indicated strong performance with R² = 0.99 under normal and 0.98 under adverse conditions, and a mean squared error (MSE) of 0.0085. Average inference latency is 110-116 ms on Jetson Nano and 210-230 ms on Raspberry Pi, confirming suitability for real-time deployment on edge hardware. The hybrid edge-cloud method enables adaptive, real-time vehicle control through IoT sensor fusion. CNN-based perception enhances prediction accuracy and operational safety under variable driving conditions. Demonstrates feasibility of deep learning deployment on low-cost edge devices for intelligent transport applications. Thus, integrating deep learning with IoT-enabled sensors on an edge-cloud platform provides a reliable and scalable pathway toward safe, adaptive, and efficient vehicle control in intelligent transportation systems.

Modeling and Subtraction (EMBS) for Video Segmentation is a photograph-processing method which makes use of a complicated probabilistic method to discover slow changes in a picture. The EMBS algorithm is designed to analyze a series of video frames with the intention to locate and subtract unobtrusive modifications in the scene. This allows for segmentation of the video records into areas of hobby that have not changed substantially, which could then be used for further analyses or operations. The algorithm employs a Gaussian mixture model (GMM) to subdivide the collection of frames into unique regions consistent with the predicted motion. A getting to know step is used to refine the GMM, allowing for more accurate segmentation. In the end, a picture-segmenting algorithm is used to detect areas of interest and to offer a refinement of the GMM. This green and sturdy algorithm is demonstrated to provide appropriate accuracy and pace for video segmentation.

Aspect computing is quickly turning into a possible answer for augmenting and streamlining the overall performance of next era community services. Edge computing gives a disbursed computing architecture that allows computation and records storage competencies to are living closer to the edge network, rather than in, valuable records middle. This enables applications to enjoy faster reaction times and reduced latency, growing the rate of facts shipping. Side computing also presents greater scalability, simplifying the combination of IoT gadgets, permitting statistics streaming from more than one asset. As networks pass to 5G, area computing will offer the important flexibility and stability to deal with big amounts of statistics visitors securely and successfully. By leveraging the power of edge computing, corporations can create an agile and resilient service structure, imparting more advantageous safety and a variety of customizable abilities that permit them to capitalize on the success of the 5 G community.

This research delves into the integration of human feedback within reinforcement learning (RL) algorithms, with a specific focus on the CartPole environment as a testbed. We present RLHFAgent, a revolutionary RL agent devised to capitalize on human guidance during training for the purpose of expediting the learning process. Through the acquisition of feedback from a human operator, RLHFAgent adapts its policy in a more efficient manner, resulting in enhanced performance when it comes to balancing the pole. Our approach involves the training of a neural network model that approximates the policy function, mapping observations to actions, and subsequently updating this model based on human feedback. By means of a series of experiments, we showcase the efficacy of RLHFAgent in learning the art of balancing the pole, as demonstrated by the consistent rise in episodic rewards and the decrease in episodic loss over the course of training episodes. These findings indicate that the incorporation of human intuition into RL algorithms can augment their ability to adapt and expedite the learning process in intricate environments. In essence, this study contributes to the ongoing endeavours aimed at bridging the gap between RL algorithms and human expertise, thereby paving the way for more efficient and effective learning strategies in both simulated and real-world scenarios.

An increasing number of sociopolitical issues, including international wars, armed conflicts, decoupling of economic powers, deglobalization, nationalism, protectionism, sanctions, tariff wars, political tensions, and other sociopolitical forces affecting global supply chains (GSCs) and supply chain management (SCM) strategies. Amidst these sociopolitical factors, the global trade landscape is undergoing a seismic shift. This paper examines the complex web of these sociopolitical aspects, their impact on SCM strategies, and their implications for GSCs. The systematic literature review approach is used for this research, which entails the following steps: selecting and assessing pertinent sources, summarizing and synthesizing the literature, examining challenges and opportunities, articulating conclusions, and pinpointing prospective directions for future research. The findings of this study reveal that these sociopolitical factors have significant impacts on multiple facets of SCM, such as sourcing strategies, supplier relationships, risk management, and overall supply chain resilience. In addition, the study presents a roadmap for businesses to effectively navigate these difficulties, which involves diversifying supply chain networks, integrating corporate sustainability strategies, and implementing environmental, social, and governance factors into business operations. The originality of this research lies in its integrated and holistic approach to the integration of sociopolitical factors into SCM theory and practice.

This research introduces an innovative approach to enhance lip reading-based text extraction and translation through the integration of a double Convolutional Neural Network (CNN) coupled with Recurrent Neural Network (RNN) architecture. The proposed model aims to leverage the strengths of both CNN and RNN to achieve superior accuracy in lip movement interpretation and subsequent text extraction. The methodology involves training the double CNN+RNN model on extensive datasets containing synchronized lip movements and corresponding linguistic expressions. The initial layers of the model utilize CNNs to effectively capture spatial features from the visual input of lip images. The extracted features are then fed into RNN layers, allowing the model to grasp temporal dependencies and contextual information crucial for accurate lip reading. The trained model showcases its proficiency in extracting textual content from spoken words, demonstrating an advanced capability to decipher nuances in lip gestures. Furthermore, the extracted text undergoes a translation process, enabling the conversion of spoken language into various target languages. This research not only contributes to the advancement of lip reading technologies but also establishes a robust foundation for real-world applications such as accessibility solutions for individuals with hearing impairments, real-time multilingual translation services, and improved communication in challenging acoustic environments. The abstract concludes with a discussion on the potential impact of the double CNN+RNN model in pushing the boundaries of human-computer interaction, emphasizing the synergy between deep learning, lip reading, and translation technologies

A large number of IoT applications have used blockchain technology in recent years. There has been a lot of buzz around blockchain oracles, which connect blockchain and off-chain data transmission. The effectiveness and safety of data collecting for oracles are greatly compromised by the vast and diverse array of devices that make up the Internet of Things (IoT). The effectiveness and reliability of the oracle system as a whole is impacted by the matching connection among data sources with oracle nodes. In order to tackle these problems, this article suggests an efficient and distributed oracle solution that is designed for the Internet of Things. It allows for the quick gathering of off-chain data in real-time. To address the heterogeneity of IoT devices, we first develop a distributed oracle design that integrates conventional and Trusted devices to enhance system scalability. Secondly, we allocate suitable nodes to jobs to improve system efficiency by determining the matching connection between nodes and their data sources, which is based on the information supplied by trusted devices about trustworthy nodes. As a second step toward making the data even more consistent, we suggest a sliding window data filtering technique. The suggested method is confirmed to be secure by doing a security analysis. The experimental findings validate the effectiveness of the suggested system in enhancing the oracle's service quality. Experiments in simulation have shown that our suggested alternative significantly improves the system's efficiency and service quality, cutting average response time by around 15% when compared to traditional methods.

Antagonistic networks are deep studying models used in gadget mastering programs that take advantage of the strengths of both generative and discriminative models. The purpose of the usage of opposed networks is to enhance the accuracy of photo category by introducing a form of regularization to the version that allows reduce over-fitting and improves generalization. On this mission, we present experiments the use of an adversarial community to improve the accuracy of a picture category community. We also discover the one-of-a-kind styles of regularization strategies that can be used to increase the version’s generalization overall performance. Subsequently, we offer a contrast of the overall performance of the class community with and without adversarial networks and talk capability applications of our effects.

Recent advances in synthetic intelligence (AI) technologies, including deep learning, have generated a hobby in growing advanced neural networks for extracting relevant facts from big unstructured datasets. In this painting, we recommend numerous enhanced modern mechanisms to improve the accuracy and speed of the latest information extraction from textual content analytics programs. In particular, we explore the usage of transfer today's semi supervised latest and reinforcement learning tactics to allow the powerful use of brand new categorized and unlabeled education statistics. We propose architectures for combining multiple neural networks and evaluate their performance on various responsibilities. Experimental results on a selection of contemporary benchmark datasets show that our cutting-edge techniques outperform strategies on duties such as text category, sentiment evaluation, and natural language understanding. The proposed more advantageous modern mechanisms may have several potential applications in finance, healthcare, and criminal investigation. Our proposed strategies can be relevant to various AI obligations and represent a critical step in cutting-edge green and correct information extraction from unstructured statistics.

Sentiment analysis is a useful tool for social media and customer analysis allowing one to glean a summary of the views of a large population regarding a particular topic. This work conducts an empirical survey of different techniques for sentiment analysis; they cover the implementation, advantages, and limitations of each of these methods and conduct an experiment to find out which of these methods is best suited for sentiment analysis in today’s scenario. Experimentation conducted on benchmark datasets and the results obtained highlight the fact that supervised learning algorithms like Support Vector Machines (SVM), Multinomial Logistic Regression (MLR), and Deep-Learning-based algorithms like Convolutional Neural Network (CNN) and Bidirectional Recursive Neural Network (RNN) show improved performances over the lexicon-based methods.

Kārakas from ancient Paninian grammar form a concise set of semantic roles that capture crucial aspect of sentence meaning pivoted on the action verb.In this paper, we propose employing a kāraka-based approach for retrieving answers in Indic question-answering systems.To study and evaluate this novel approach, empirical experiments are conducted over large benchmark corpora in Hindi and Marathi.The results obtained demonstrate the effectiveness of the proposed method.Additionally, we explore the varying impact of two approaches for extracting kārakas.The literature surveyed and experiments conducted encourage hope that kāraka annotation can improve communication with machines using natural languages, particularly in low-resource languages.

Over the past few years, research interest in the sub-domain of question answering has tremendously increased. Yet, most of the work on QA and more generally, on natural language processing has been predominantly limited to the English language. In contrast, with each passing year, the number of people with access to the internet is exponentially increasing, especially those residing in South Asian countries whose primary language is not English. With this in mind, the survey’s aim is to recognize, review and analyze the various question-answering datasets that exist for resource-scare Indic languages such as Hindi, Urdu, Tamil, and Marathi. It also intends to shed light on the state-of-the-art of Indic question-answering itself, in terms of methods used, best-performing models, and evaluation metrics. The review also includes multilingual benchmarks which have been recently published.

The Encapsulating Safety Payload (ESP) is a necessary part of the IPsec suite of protocols, imparting a critical layer of security within a Wi-Fi community. This paper aims to evaluate ESP and its software in wireless community protection comprehensively. ESP capabilities as a secured payload encapsulating data via imparting integrity, authentication, and confidentiality. It commonly accomplishes this by using a mixture of cryptography and tunneling to shield the contents of the packets from alteration, eavesdropping, and unauthorized get entry. Moreover, this paper outlines the distinctive ESP modes and cipher suites, presenting how they may be used to defend data being despatched over a Wi-Fi community. ESP additionally affords numerous safety enhancements, such as protection from replay assaults, packet length control, and critical negotiation. Furthermore, this paper explores the challenges associated with implementing and deploying ESP in wireless networks. Subsequently, the paper will pay attention to the numerous overall performance issues for ESP utilization by analyzing the effect of latency, throughput, and power intake on ESP-enabled Wi-Fi networks.

The increasing prevalence of e-commerce has empowered consumers with vast choices and opportunities for online shopping. This research paper focuses on two essential aspects of online shopping: price comparison and sentiment analysis of product reviews. The paper presents a methodology for scraping product prices from multiple e-commerce websites and conducting sentiment analysis on the corresponding product reviews. The findings of this research have significant implications for both consumers and e-commerce businesses. Consumers can leverage price comparison data to identify the most cost-effective platforms for their desired products, while sentiment analysis enables them to assess the overall satisfaction levels of other customers. E-commerce businesses can utilize these insights to optimize pricing strategies, identify areas for improvement, and enhance customer experiences. Performance analysis of Support Vector Machine, Logistic Regression, VADER Lexicon and SentiWordNet Lexicon is also done.

Data-driven models function admirably in solving real-world problems. However, obtaining relevant data is difficult. Also, sometimes more diverse data is needed to identify the limitations of trained Machine Learning models. Creating such data samples based on earlier known metadata is a common practice. However, this process can induce bias in the dataset unknowingly. Generative Adversarial Networks (GAN) based data generation models generate more data based on initial data distribution. Thus, data generation models may reflect bias in the generated synthetic data. In this study, the authors have proposed an interactive synthetic data generation Graphical User Interface (GUI) tool. The tool is equipped with Bias detection and mitigation algorithms which will notify users about the pre-existing bias and provide methods to mitigate it. Similarly, this tool can be used to evaluate synthetic data generated using GAN-based models against fairness metrics. The authors have found that Learning Fair Representation (LFR) bias mitigation method has performed 62% & 17.5% better than Prejudice remover and Disparate impact remover for German Credit & Adult original datasets. These results were concluded based on bias detection metrics such as Statistical Parity Difference (SPD) and Disparate Impact (DI). The proposed data generation tool used with LFR method can reduced SPD metric by 93% on original German Credit data. The authors conclude that both original and synthetic datasets had a bias. Therefore, the fairness level of any dataset should be checked vigilantly.

It is vital that any type of text can be portrayed in a machine-readable language so that it can be processed and interpreted. Text embeddings, which allow a machine to interpret any type of text by mathematically expressing it, naturally minimise this work. To extract the concept of interdependence across words or sentences, the degree of connection, synonym recognition, idea segmentation, selectional preferences, and analogies, text embedding are very useful. Using closest neighbor lookup of vectors, a vector representation of a text in a semantic vector space may be utilized for answer retrieval according to the vectors’ similarity and proximity. Traditional QA systems require a huge number of high-quality language-specific resources to perform multiple activities. The precision of several embedding techniques presently supporting Marathi, an Indo-Aryan language, is explored in this research. The performance of several sentence embedding algorithms is evaluated to select the most relevant and related answer text within the corpus given a natural language query in Marathi. Pragmatic tests are performed across supervised machine learning models to perform categorization and sentence embeddings in order to evaluate and compare the veracity of various embedding techniques. The proposed research findings concentrate on identifying the most effective embedding approaches from the restricted resources available for the Marathi language, laying the groundwork for future study in this language.

AI-Powered tools and technologies have proven to be successful in solving a variety of business problems including sales and marketing optimization. The social media marketing campaign is something widely used today but requires deep domain expertise and human efforts and hence an expensive approach. This problem is most prevalent in Small and Medium Enterprises (SMEs) where the lack of a Cost-effective solution hinders their ability to leverage the power of social media outlets for revenue and mind share growth. In this paper, we propose a system and methods for Automatic marketing Campaign generation using AI models with a Data-driven approach. Kaggle dataset of supermarket analysis was used for experimentation and Natural Language Generation (NLG) technology was used with canned queries for Text Generation, lastly, images were retrieved from the database using Deep Learning-Based Object detection to make the campaign more visually appealing to the clients. This methodology reduces the human efforts required earlier, is more efficient, and helps improve the marketing campaign’s reach by further publishing it on social media. This approach achieved satisfactory good results and responses after validation by the industry experts.

Answer sentence selection is an important sub-task in Question Answering (QA) that determines the correct answer sentence from a passage. This task can naturally be reduced to the semantic text similarity problem between question and answer candidate. In this work, we investigate the significance of various similarity measures for the answer sentence selection task in Hindi an Indo-Aryan language. Karaka relations is the core of dependency annotation scheme used for Hindi and are crucial to syntactico-semantic analysis of the sentence. We investigate this, and compare them to other, hitherto known measures. To investigate and compare the utility of various measures, we develop a test-bench over a benchmark Hindi and English multilingual QA corpus for comparison, making two tool-chains and designing empirical experiments across combinations of similarity measures, sentence embedding schemes, and supervised machine learning models for classification. Combining Karaka relations with different similarity measures shows significant performance improvement for sentence selection task, suggesting them as potentially a semantic similarity measure. Moreover, our results give us confidence that refinement of Karaka relations extraction to optimal quality will reduce the need for availability of large pre-trained language models.

The widespread use of image-based memes on socioeconomic or political issues has witnessed a booming effect unparallel to any form of media in the recent years. The ability to go viral on social media in seconds and the popularity of memes on online platforms give a wide scope and pathway for research as it will help in understanding the usage patterns of the public and in turn be used for analyzing their sentiment toward a specific topic/event. In this paper, initially gap analysis on the features used for sentiment extraction on memes is presented. Exploring the correlation of image based and textual features, this paper gives a novel approach (correlating the facial features along with the text in the meme itself) for the extraction of sentiment from image-based memes. This paper also addresses the challenges faced in this relatively new area of sentiment extraction on memes. Finally, this paper concludes with insightful results.

In the age of big data analytics, text mining on large sets of digital textual data does not suffice all the analytic purposes. Prediction on unstructured texts and interlinking of the information in various domains like strategic, political, medical, financial, etc. is very pertinent to the users seeking analytics beyond retrieval. Along with analytics and pattern recognition from the textual data, there is a need to formulate this data and explore the possibility of predicting future event(s). Event prediction can best be defined as the domain of predicting the occurrence of an event from the textual data. This paper spans over two main sections of surveying technical literature and existing tools/technologies functioning in the domain of prediction and analytics based on unstructured text. The survey also highlights fundamental research gaps from the reviewed literature. A systematic comparison of different technical approaches has been listed in a tabular form. The gap analysis provides future scope of more optimized algorithms for textual event prediction.