Convolutional neural network (CNN) accelerators powered by FPGA have garnered a lot of interest lately. This is mainly because they provide a higher degree of energy efficiency when compared to Graphics processing uni...
详细信息
In efforts to better accommodate users, numerous researchers have endeavored to model customer behavior, seeking to comprehend how they interact with diverse items within online platforms. This exploration has given r...
详细信息
In efforts to better accommodate users, numerous researchers have endeavored to model customer behavior, seeking to comprehend how they interact with diverse items within online platforms. This exploration has given rise to recommendation systems, which utilize customer similarity with other customers or customer-item interactions to suggest new items based on the existing item catalog. Since these systems primarily focus on enhancing customer experiences, they overlook providing insights to sellers that could help refine the aesthetics of their items and increase their customer coverage. In this study, we go beyond customer recommendations to propose a novel approach: suggesting aesthetic feedback to sellers in the form of refined item images informed by customer-item interactions learned by a recommender system from multiple consumers. These images could serve as guidance for sellers to adapt existing items to meet the dynamic preferences of multiple users simultaneously. To evaluate the effectiveness of our method, we design experiments showcasing how changing the number of consumers and the class of item image used affect the change in preference score. Through these experiments, we found that our methodology outperforms previous approaches by generating distinct, realistic images with user preference higher by 16.7%, thus bridging the gap between customer-centric recommendations and seller-oriented feedback. Copyright 2025 Kumar et al. Distributed under Creative Commons CC-BY 4.0
Language is crucial for expressing thoughts, ideas, and feelings, but computers typically understand structured language, processing it to generate meaningful results. However, many fields, like the legal domain, use ...
详细信息
Language is crucial for expressing thoughts, ideas, and feelings, but computers typically understand structured language, processing it to generate meaningful results. However, many fields, like the legal domain, use unstructured or semi-structured language, which lacks specific syntax and context-dependent semantics. Text summarization was once challenging, but with the advent and widespread use of Large Language Models (LLMs), it has become more efficient. Legal briefs are concise summaries of extensive judgments requiring legal decision-making. This research investigates legal text summarization, examining if current symbolic or non-symbolic AI models can effectively generate legal case briefs. The study uses the Rhetorical Roles (RRs) generated by the BUILDNyAI RR model and introduces a novel Rhetorical Role Relatedness (RRR) framework. This research had tried to augment the legal reasoning process that will help a legal profession to drafting legal briefs. This framework successfully creates legal briefs closely matching the golden standards using RR Similarity Scores and the LexRank summarizer, achieving higher ROUGE scores (ROUGE-1, ROUGE-2, ROUGE-L-Sum) compared to the BUILDNyAI Summarizer model
The Sign Language Recognition System has been designed to capture video input, process it to detect hand gestures, and translate these gestures into readable text. The project consists of several key components and st...
详细信息
ISBN:
(数字)9798331523923
ISBN:
(纸本)9798331523930
The Sign Language Recognition System has been designed to capture video input, process it to detect hand gestures, and translate these gestures into readable text. The project consists of several key components and steps: Video Processing: Using OpenCV, the system captures frames from the video input. MediaPipe processes these frames to detect and track hand landmarks in real time. OpenCV capabilities allow for efficient frame extraction and basic image processing tasks such as resizing and normalization. Hand Detection and Tracking: MediaPipe pre-trained models identify and track hand movements within the video frames. The accurate detection and tracking of the hand movements are critical for the subsequent recognition of the sign language gestures. Sign Language Recognition: The core system is the deep learning model, trained using the TensorFlow and Keras on a dataset of sign language gestures. The model learns to classify the detected hand movements into corresponding sign language characters or words. Convolutional Neural Networks (CNNs) are typically used for task due to their effectiveness in image recognition tasks. Text Display: Once the system recognizes the signs, it converts them into text and displays the output. This can be done through a console output or a graphical user interface (GUI) built with Tkinter. The GUI provides a user friendly experience, allowing users to see the translated text in real-time.
K-means clustering is a fundamental data mining technique. It heavily relies on parameter optimization (number of clusters, initial centers, and distance measures) for accurate and meaningful results. This study addre...
详细信息
We study the time-refraction of optical waves as they propagate through a time-varying slab and find the frequency shift associated with the abrupt change in the refractive index. We discover a geometrical mechanism r...
详细信息
Anomaly detection and classification play a vital role in maintaining public safety and security. An automated system for anomaly detection and classification can reduce human leverage, cost, and time. We propose a tw...
详细信息
ISBN:
(数字)9798350357509
ISBN:
(纸本)9798350357516
Anomaly detection and classification play a vital role in maintaining public safety and security. An automated system for anomaly detection and classification can reduce human leverage, cost, and time. We propose a two-stage classifier pipelined within a single model for anomaly detection and classification. The first stage of the two-stage classifier is a Convolutional Neural Network(CNN) based binary classifier that determines whether an event is anomalous or normal. Based on the output of the first classifier, if the event is found anomalous, then it goes to the second stage of our single-pipelined model. The second stage classifier is a Vision Transformer (ViT) based architecture that further classifies an anomalous event into specific anomaly categories. This research utilized the UCF Crime dataset. Being quite large, it requires a significant amount of computational resources and time for processing. We also proposed a keyframe extraction algorithm to reduce the computational cost and time. The proposed keyframe extraction algorithm identifies and selects only relevant frames from videos and discards the redundant and irrelevant frames. The proposed methodology combines Convolution Neural Network (CNN) and Vision Transformer (ViT) for spatial-temporal feature extraction from a complex scenario and classify them. The proposed model achieves 98% accuracy for binary classification modules and 95% accuracy for multi-class classification. Furthermore, the proposed keyframe extraction algorithm significantly reduces the processing time and computational resources. For each videos, it requires only 20ms processing time. The outcome of the proposed model suggests that it can outperform traditional methods for anomaly detection and classification. However, a highly correlated and vast amount of data creates problems like overfitting and increases the complexity of the model.
Crowdfunding has gained popularity as an innovative approach to raise capital for startups, social initiatives, and creative ventures. However, traditional centralized crowdfunding platforms often face critical challe...
详细信息
ISBN:
(数字)9798331512248
ISBN:
(纸本)9798331512255
Crowdfunding has gained popularity as an innovative approach to raise capital for startups, social initiatives, and creative ventures. However, traditional centralized crowdfunding platforms often face critical challenges such as lack of transparency, risk of fraud, and high intermediary costs. To address these limitations, this paper proposes a decentralized crowdfunding platform built on the Ethereum blockchain, leveraging smart contract technology to ensure transparency, accountability, and security. The primary objective of this work is to design a trustworthy and automated system where funders and fundraisers can interact directly without relying on third-party intermediaries. The platform enforces pre-defined conditions for fund distribution, allowing profit-sharing and fund release mechanisms to be executed autonomously through smart contracts. This enhances both the reliability and precision of transactions. Funders have real-time visibility into project progress, improving their confidence and decision-making. The proposed solution also reduces operational costs by automating core functionalities like funding, profit allocation, and capital release. All interactions are securely recorded on the blockchain, ensuring auditability and eliminating manipulation. This paper presents the system's architecture, implementation strategy, and evaluation results, demonstrating the effectiveness of the blockchain-based model in overcoming the shortcomings of centralized platforms. The proposed platform offers a more secure, efficient, and transparent alternative for the crowdfunding ecosystem.
Plenty of different diagnosing methods have been extensively utilized to identify diabetes accurately;however, an absolutely precise and definitive diagnosis has not yet been attained. Within the context of this resea...
详细信息
ISBN:
(纸本)9783031828805
Plenty of different diagnosing methods have been extensively utilized to identify diabetes accurately;however, an absolutely precise and definitive diagnosis has not yet been attained. Within the context of this research, our primary objective is to leverage the cutting-edge capabilities of Artificial Intelligence (AI) coupled with OpenCV to assist medical professionals, thereby minimizing the rate of misdiagnosis. Specifically, we harness the power of AI to effectively classify diverse images portraying patients afflicted with Non-Proliferative Diabetic Retinopathy (NPDR), with the ultimate goal of determining the severity level at which they are situated. In conjunction with this, Python, with OpenCV, has a crucial role in extracting pertinent features that may be discernible within the given images. Our methodology involves the collection and preprocessing of the Eye PACS Dataset on Kaggle, followed by feature extraction and model training using some machine learning algorithms, including convolutional Neural Network CNN, decision trees, support vector machines SVM, and neural networks. OpenCV is utilized for image processing tasks, enhancing the feature extraction process, certain individual features present within the images are precluded from being considered as contributing factors in the classification process. Some of these features include but not limited to, the measurement of the luminous blobs which are present in the image, the specific area of existence of red lesion. The evaluation of the models includes the analysis of their performance based on the goal of the prediction task, specifically decimal-based accuracy, precision, recall, and F1-score. This research employs a wide-ranging dataset embracing low, medium and high level of image severity. At last, after lots of simulation, it came to a conclusion that the CNN increases its level of classification accuracy up to 98%. These findings show that the proposed application of AI improves the accuracy
暂无评论