Adversarially Resilient and Lightweight AI for Real-Time Fraud Detection in Edge-Enabled Financial Networks

The digitization of the financial industry through technologies such as FinTech, mobile payments, and decentralized finance has widened the attack surface and increased the need for immediate transaction processing. AI and deep learning are now central to detecting complex, non-linear fraud patterns, but deploying these models is held back by three challenges. First, speed and scale: financial transactions must clear in milliseconds, yet heavy models such as deep learning ensembles add latency that real-time systems cannot absorb, especially at the edge on mobile phones and IoT payment terminals. Second, adversarial AI: fraudsters increasingly craft inputs designed to trick and evade detection models. Third, data and computational limits: privacy regulations and network constraints often rule out pooling all data in one place for training, while running large models at the edge forces trade-offs in computation and energy use.This Research Topic brings together studies that address these challenges jointly. We want to advance AI models that are accurate, resilient to attacks, resource-efficient in memory, computation, and energy, and deployable at the edge. The wider goal is to move financial AI from an opaque system toward a dependable, deployment-ready technology.We welcome original research, reviews, and case studies. Areas of interest include, but are not limited to:Adversarial robustness of financial AI: • Defence methods designed for transaction data and temporal or time-series sequences • Development and evaluation of models against adversarial threats specific to financial contexts • Empirical studies on the vulnerability of current fraud detection models to different attack typesLightweight and resource-efficient deep learning architectures: • Architecture optimization methods applied to fraud detection models • Development of efficient deep learning models • Evaluation of the trade-offs between model complexity, detection accuracy, and inference speed on resource-constrained edge and mobile devicesResource-constrained and federated learning frameworks: • Privacy-preserving learning methods that improve robustness and personalization across financial networks • Combining federated learning with generative AI to synthesize privacy-preserving data for training in data-scarce settings • Algorithms for robust and fair collaborative learning across heterogeneous, non-IID data from different financial institutions Topic editors: Prashant Upadhyay, RUPA RANI, PAWAN KUMAR GOEL Source: https://www.frontiersin.org/research-topics/84018/adversarially-resilient-and-lightweight-ai-for-real-time-fraud-detection-in-edge-enabled-financial-networks