AURNA: Asymmetric Uncertainty-Responsive Neural Architecture for Granular Fraud Detection

Light field saliency detection is becoming of increasing interest in recent years due to the significant improvements in challenging scenes by using abundant light field cues. However, high dimension of light field data poses computation-intensive and memory-intensive challenges, and light field data access is far less ubiquitous as RGB data. These may severely impede practical applications of light field saliency detection. In this paper, we introduce an asymmetrical two-stream architecture inspired by knowledge distillation to confront these challenges. First, we design a teacher network to learn to exploit focal slices for higher requirements on desktop computers and meanwhile transfer comprehensive focusness knowledge to the student network. Our teacher network is achieved relying on two tailor-made modules, namely multi-focusness recruiting module (MFRM) and multi-focusness screening module (MFSM), respectively. Second, we propose two distillation schemes to train a student network towards memory and computation efficiency while ensuring the performance. The proposed distillation schemes ensure better absorption of focusness knowledge and enable the student to replace the focal slices with a single RGB image in an user-friendly way. We conduct the experiments on three benchmark datasets and demonstrate that our teacher network achieves state-of-the-arts performance and student network (ResNet18) achieves Top-1 accuracies on HFUT-LFSD dataset and Top-4 on DUT-LFSD, which tremendously minimizes the model size by 56% and boosts the Frame Per Second (FPS) by 159%, compared with the best performing method.

Most existing RGB-D saliency detection methods adopt symmetric two-stream architectures for learning discriminative RGB and depth representations. In fact, there is another level of ambiguity that is often overlooked: if RGB and depth data are necessary to fit into the same network. In this paper, we propose an asymmetric two-stream architecture taking account of the inherent differences between RGB and depth data for saliency detection. First, we design a flow ladder module (FLM) for the RGB stream to fully extract global and local information while maintaining the saliency details. This is achieved by constructing four detail-transfer branches, each of which preserves the detail information and receives global location information from representations of other vertical parallel branches in an evolutionary way. Second, we propose a novel depth attention module (DAM) to ensure depth features with high discriminative power in location and spatial structure being effectively utilized when combined with RGB features in challenging scenes. The depth features can also discriminatively guide the RGB features via our proposed DAM to precisely locate the salient objects. Extensive experiments demonstrate that our method achieves superior performance over 13 state-of-the-art RGB-D approaches on the 7 datasets. Our code will be publicly available.KeywordsSaliency detectionFlow ladderDepth attention

Financial fraud detection is a critical application area within the broader domains of cybersecurity and intelligent financial analytics. With the growing volume and complexity of digital transactions, the traditional rule-based and shallow learning models often fall short in detecting sophisticated fraud patterns. This study addresses the challenge of accurately identifying fraudulent financial activities, especially in highly imbalanced datasets where fraud instances are rare and often masked by legitimate behavior. The existing models also lack interpretability, limiting their utility in regulated financial environments. Experiments were conducted on three benchmark datasets: IEEE-CIS Fraud Detection, European Credit Card Transactions, and PaySim Mobile Money Simulation, each representing diverse transaction behaviors and data distributions. The proposed methodology integrates a transformer-based encoder, multi-teacher knowledge distillation, and a symbolic belief–desire–intention (BDI) reasoning layer to combine deep feature extraction with interpretable decision making. The novelty of this work lies in the incorporation of cognitive symbolic reasoning into a high-performance learning architecture for fraud detection. The performance was assessed using key metrics, including the F1-score, AUC, precision, recall, inference time, and model size. Results show that the proposed transformer–BDI model outperformed traditional and state-of-the-art baselines across all datasets, achieving improved fraud detection accuracy and interpretability while remaining computationally efficient for real-time deployment.

We present a unified multi-task learning architecture for fast and accurate pedestrian detection. Different from existing methods which often focus on either a new loss function or architecture, we propose an improved multi-task convolutional neural network learning architecture to effectively and efficiently interfuse the task of pedestrian detection and semantic segmentation. To achieve this, we integrate a lightweight semantic segmentation branch to Faster R-CNN detection framework that enables end-to-end hard parameter sharing in order to boost the detection performance and maintain computational efficiency as follows. Firstly, a Semantic Segmentation to Feature Module (SS2FM) refines the convolutional features in RPN stage by integrating the features generated from the semantic segmentation branch. Secondly, a Semantic Segmentation to Confidence Module (SS2CM) refines the classification confidence in RPN stage by fusing it with the semantic segmentation confidence. We also introduce an effective anchor matching point transform to alleviate the problem of feature misalignment for heavily occluded pedestrians. The proposed unified multi-task learning architecture lends itself well to more robust pedestrian detection in diverse scenarios with negligible computation overhead. In addition, the proposed architecture can achieve high detection performance with low resolution input images, which significantly reduces the computational complexity. Experiment results on CityPersons and Caltech datasets show that our method is the fastest among all state-of-the-art pedestrian detection methods while exhibiting competitive detection performance.

The increasing complexity of fraud in large enterprises, particularly in financial and transactional ecosystems, necessitates a scalable, agile, and distributed detection approach. Traditional centralized fraud detection architectures struggle to keep pace with the real-time requirements, data silos, and domain-specific fraud patterns that arise across organizational units. This paper explores the application of Data Mesh architecture as a decentralized and domain-oriented paradigm for enhancing fraud detection in large enterprises. Data Mesh shifts the ownership of data from centralized teams to domain-specific teams, treating data as a product and enabling better scalability, autonomy, and responsiveness. We propose a Data Mesh-based fraud detection model wherein each business domain—such as sales, finance, customer relations, and operations—operates as a semi-autonomous node capable of detecting fraud patterns locally while contributing to an enterprise-wide fraud intelligence network. Leveraging a federated governance model, the architecture facilitates standardized yet decentralized policy enforcement, model deployment, and cross-domain collaboration. This paper examines how data product thinking, domain-driven design, self-serve data platforms, and federated computational governance work together to create a resilient and adaptable architecture for fraud detection. The proposed methodology utilizes distributed anomaly detection algorithms, local event-driven stream processing (e.g., Apache Kafka and Flink), and inter-domain feedback loops for continuous model retraining and behavior correlation. Experimental simulations conducted on synthetic multi-domain enterprise data reveal improved time-to-detection, reduced false positives, and enhanced fraud detection in low-signal data scenarios compared to centralized models. Furthermore, the architecture demonstrates superior scalability and flexibility when integrating new domains and updating detection logic. The findings of this research indicate that Data Mesh not only democratizes access to fraud-related data but also enhances detection capabilities by aligning technical solutions with organizational complexity. This paper contributes to the growing body of decentralized AI applications in enterprises and offers actionable design patterns for implementing domain-centric fraud analytics in large organizations. Future work includes extending this architecture to incorporate privacy-preserving technologies such as federated learning and exploring its applicability in regulatory compliance frameworks

The exponential growth of data in recent years has created significant challenges in fraud detection. Fraudulent activities are increasingly widespread across sectors, such as banking, web networks, health insurance, and telecommunications. This trend highlights a growing need for big data technologies such as Hadoop, Spark, Storm, and HBase to enable real-time detection and analysis of data fraud. This study aims to enhance understanding of the fraud classifications and their spread in various sectors. Fraud detection involves analyzing data and developing machine learning (ML) models or traditional rule-based systems to identify abnormal activities as they occur. The analysis in this paper examines both the advantages and limitations of these solutions, particularly regarding scalability and performance. This paper evaluates the methods and big data tools used in fraud detection and prevention through a comprehensive literature review, emphasizing the implementation challenges. This review discusses existing solutions, operational environments, and the ML algorithms and traditional rules employed. The main objective of this study is to address these challenges by proposing an innovative architecture that equips organizations with the latest knowledge and methodologies in big data technologies for real-time fraud detection and prevention.

This article presents a comprehensive architectural framework for implementing secure multi-tenant FinTech platforms that leverage artificial intelligence for real-time fraud detection while maintaining stringent regulatory compliance and data security standards. The proposed architecture addresses the complex challenges of deploying AI-driven financial services across shared infrastructure environments through innovative approaches, including containerized database sharding, attribute-based access control systems, and secure enclave computation technologies. The framework integrates Apache Kafka and Apache Flink streaming platforms to enable high-velocity transaction processing with end-to-end encryption protocols, ensuring data isolation between tenants while supporting cross-tenant analytical capabilities essential for effective machine learning model training and inference. Advanced AI model implementations incorporate ensemble learning techniques for credit risk assessment and deep learning architectures for fraud detection, utilizing dynamic threshold management systems and automated response frameworks to optimize performance across diverse financial scenarios. The architecture's compliance framework addresses Payment Card Industry Data Security Standard, General Data Protection Regulation, and Sarbanes-Oxley Act requirements through comprehensive audit trails, immutable compliance records, and automated policy enforcement mechanisms that adapt dynamically to changing regulatory landscapes across multiple jurisdictions.

This article examines the transformative impact of artificial intelligence on fraud detection and compliance monitoring in the financial sector. The article investigates how advanced machine learning techniques, particularly Isolation Forest algorithms and Graph Neural Networks, enable financial institutions to identify suspicious patterns and anomalies in transaction data that traditional rule-based systems often miss. The article presents a comprehensive framework for implementing AI-driven fraud detection systems that balance detection accuracy with computational efficiency while addressing the challenges of model explainability and regulatory compliance. Through multiple case studies across banking, insurance, and cross-border transactions, we demonstrate how these technologies significantly enhance detection capabilities while reducing false positives. The article also explores the ethical and regulatory considerations surrounding AI deployment in financial compliance, proposing guidelines for responsible implementation that maintain privacy protections while satisfying regulatory requirements. The article suggests that properly implemented AI methodologies represent a substantial advancement in the financial industry's ability to combat increasingly sophisticated fraud schemes while streamlining compliance processes.

This article proposes a hybrid and modular architecture for fraud detection that integrates both offline and online machine learning models to address challenges in dynamic financial transaction environments. The framework combines high-performance offline models, including XGBoost, LightGBM, and deep neural networks, with lightweight and adaptive online learners, such as Hoeffding Trees and Adaptive Random Forests, enabling accurate detection in both historical datasets and real-time streaming transactions. A key methodological contribution lies in balancing predictive performance, responsiveness, and interpretability, achieved through a weighted risk scoring mechanism and a unified cost-sensitive evaluation framework that aligns technical metrics with tangible financial impacts. The architecture emphasizes modularity and scalability, facilitating continuous adaptation via concept drift detection and feedback-driven retraining. Its implementation in a containerized, open-source environment ensures reproducibility, robustness, and seamless deployment in production-grade financial ecosystems, even under high-volume transactional loads. The proposed system effectively bridges the gap between advanced machine learning research and operational requirements, providing a flexible, interpretable, and operationally viable solution for modern fraud detection.Furthermore, this study consolidates previous work by the author on intelligent fraud detection systems, extending prior contributions in model selection, AI interpretability, and economic evaluation of false positives in banking contexts. Future research directions include integrating graph-based relational features for network fraud detection, applying reinforcement learning for adaptive decision optimization, and employing federated learning techniques to enhance data privacy across institutions. Overall, the proposed framework represents a scalable, transparent, and adaptive approach that evolves alongside emerging fraud strategies, delivering a deployable system with practical and financial relevance.

Financial fraud detection remains a critical challenge due to the dynamic and adversarial nature of fraudulent behavior. As fraudsters evolve their tactics, detection systems must combine robustness, adaptability, and precision. This study presents a hybrid architecture for credit card fraud detection that integrates a Mixture of Experts (MoE) framework with Recurrent Neural Networks (RNNs), Transformer encoders, and Autoencoders. Each expert module contributes a specialized capability: RNNs capture sequential behavior, Transformers extract high-order feature interactions, and Autoencoders detect anomalies through reconstruction loss. The MoE framework dynamically assigns predictive responsibility among the experts, enabling adaptive and context-sensitive decision-making. Trained on a high-fidelity synthetic dataset that simulates real-world transaction patterns and fraud typologies, the hybrid model achieved 98.7% accuracy, 94.3% precision, and 91.5% recall, outperforming standalone models and classical machine learning baselines. The Autoencoder component significantly enhanced the system’s ability to identify emerging fraud strategies and atypical behaviors. Beyond technical performance, the model contributes to broader efforts in financial governance and crime prevention. It supports regulatory compliance with Anti-Money Laundering (AML) and Know Your Customer (KYC) protocols and aligns with routine activity theory by operationalizing AI as a capable guardian within financial ecosystems. The proposed hybrid system offers a scalable, modular, and regulation-aware approach to detecting increasingly sophisticated fraud patterns, contributing both to the advancement of intelligent systems and to the strengthening of institutional fraud defense infrastructures.

This paper describes our system (MIC-CIS) details and results of participation in the fine grained propaganda detection shared task 2019. To address the tasks of sentence (SLC) and fragment level (FLC) propaganda detection, we explore different neural architectures (e.g., CNN, LSTM-CRF and BERT) and extract linguistic (e.g., part-of-speech, named entity, readability, sentiment, emotion, etc.), layout and topical features. Specifically, we have designed multi-granularity and multi-tasking neural architectures to jointly perform both the sentence and fragment level propaganda detection. Additionally, we investigate different ensemble schemes such as majority-voting, relax-voting, etc. to boost overall system performance. Compared to the other participating systems, our submissions are ranked 3rd and 4th in FLC and SLC tasks, respectively.

This paper offers a detailed discussion of a large–scale, real-time architecture for fraud detection specifically for use in financial organizations to combat fraudulent activities in online transactions. The proposed system in this paper uses big data capabilities and a multi-stage fraud detection pipeline to detect and combat fraudulent activities efficiently. The implemented technologies include Apache Kafka, KSQL, and Spark alongside Isolation Forest algorithm for behavioral analysis of customer transactions. The presentation of the fraud detection pipeline as a series of layers exemplifies how a transaction goes through an exacting sequence of detection algorithms with very little delay and maximum precision. Verification by simulation uses the dataset of more than one hundred million Internet transactions, the performance indicators of which are a rather high F1-score of 91% and a recall rate of 97%. The results stress the advantage of the proposed methodology over conventional techniques, suggesting the possibility of real-time fraud identification. Furthermore, the paper outlines research directions where future work should focus, such as reducing computational complexity and applying deep learning solutions to enhance the detection of new types of fraud.

One of the most important factors in a purchasing decision nowadays is the evaluation of comments online. Businesses or individuals use deceptive comments to mislead people for the sake of economic gain and thus hurt the benefit of the customers and the welfare of society. In this research, we propose a rule-based artificial intelligence (AI) machine learning (ML) architecture for online review fraud detection. The proposed methodology features an AI and ML hybrid architecture, where ML refers to the common well-developed machine learning models, and the AI part features a rules-based controller that prioritizes and customizes the fraud detection rules based on human intelligence to improve the accuracy of the result and computational efficiency.

Asymmetric augmented paradigm-based graph neural architecture search

Vibration measurement serves as the basis for structural damage detection. To detect damage, vibration measurement and frequency estimation through image sequence analysis continue to receive increasing attention. In this work, we demonstrate that structural damage prediction can be achieved using a deep learning neural network architecture. In this paper, we seek to learn and see the structural damage directly from videos using deep convolutional neural networks (CNN). The key idea is to use each pixel of an image taken from a digital camera, extracting the spatiotemporal information, like a sensor to capture the modal frequencies of a vibrating structure. We develop attention-based architecture to detect subtle signals from a specific source to visualize high resolution of dynamic properties of the structures to infer existing structural damage. We first extract the high discriminative features of video frames using the CNN. Then we leverage conv-long short-term memory (ConvLSTM) with the extracted features as inputs to capture the temporal dynamics in videos. The attention mechanisms are embedded in the network to ensure the model learns to focus selectively on the dynamic frames across the video clips. Our computer vision-based deep learning model takes the video of a vibrating structure as input and outputs about the health of the structure. We demonstrate, using reliable empirical results, the proposed model is efficient, autonomous, and accurate. The proposed method is verified using a few laboratory experiments. Our experimental results demonstrate that the proposed method can achieve acceptable prediction accuracy even.