2786 papers
This paper demonstrates that TinyML models on constrained IoT devices are vulnerable to stealthy voltage and electromagnetic fault injection attacks that can degrade accuracy or force predictable outputs, and proposes Randomized Self-Reduction and majority voting as effective countermeasures to enhance their robustness.
This paper demonstrates that centered two's-complement arithmetic masking is inherently more vulnerable to side-channel attacks than unsigned masking, providing exact finite-noise optimal attacks, a new spectral leakage law, and capacity bounds that prove unsigned storage minimizes leakage for most practical moduli.
This paper presents the design, implementation, and security analysis of a dynamic session key-driven symmetric substitution cipher that resists standard attacks through position integrity and key-conditional changes, while also evaluating its performance using machine learning models to establish it as an efficient alternative for real-time encryption.
This paper introduces EGAMA-MF, a middleware framework that enhances memory-forensic malware triage by gating out structurally valid but forensically inadmissible feature vectors caused by extraction failures, thereby preventing false negatives and ensuring evidence integrity across diverse fault scenarios.
This study compares CNN-LSTM and Graph Neural Network models for credit card fraud detection on the Kaggle dataset, finding that while the CNN-LSTM outperforms the GNN due to the latter's inability to leverage anonymized relational features, the results underscore the critical importance of threshold tuning, class imbalance handling, and model selection for practical deployment.
This study reveals that large language models frequently fail to honor patient requests to withhold sensitive information in clinical documentation, disclosing such data in over a third of cases to unrelated recipients, but demonstrates that implementing recipient-conditioned routing workflows can significantly mitigate this risk without compromising necessary information sharing.
This paper proposes a strongly connected graph-based digital chaotic system constructed via an inverse approach to eliminate finite precision effects and ensure rigorous chaos properties, which is then applied to a bidirectional-diffusion image encryption algorithm that demonstrates robust security against statistical and differential attacks.
This study proposes a novel human-centric hybrid framework that integrates decentralized Multi-Agent Systems with interpretable machine learning models to enhance the accuracy, transparency, and efficiency of automated warranty fraud detection in supply chains.
This paper proposes a two-stage fine-tuning framework that enhances a multi-modal large model's facial action unit recognition and emotion classification capabilities through a question-answer dataset and decision-level fusion with a specialized small model, thereby significantly improving the accuracy and interpretability of facial expression recognition for subtle and low-intensity expressions.
This paper proposes a systematic four-phase methodology for designing custom Fourier-based SVM kernels that incorporate directional slope information, successfully reducing false positives and improving leak detection accuracy in industrial gas facilities compared to standard kernel methods.