High-throughput steganographic coding remains difficult because syndrome-trellis codes (STC) incur high computational cost at practical security settings, while steganographic polar codes retain strong serial dependencies in their successive decoding schedule. This paper proposes CuSPC, a GPU-oriented steganographic polar coding architecture that restructures simplified successive cancellation (SSC) execution, introduces an ordinary-node (Ord) design for short residual subtrees, and combines inter-cover parallelism with interleaved global-memory access. The implementation maps intra-node, inter-node, and inter-cover parallelism to the CUDA execution model while preserving the distortion-minimizing formulation of polar steganographic coding. Experiments show that CuSPC reaches up to approximately 39 Mb/s across the tested devices and improves throughput by about 30 times over the CPU serial baseline, while keeping the coding loss below 2
Protecting sensitive biomedical data is a critical security challenge in decentralized healthcare environments without a trusted authority. Electroencephalography (EEG) signals constitute highly sensitive biometric data used for neurological disorder diagnosis, making privacy-preserving analysis essential. Local differential privacy (LDP) addresses this challenge by perturbing data at the source; however, selecting an appropriate privacy budget ε remains nontrivial, as it directly governs the privacy–utility trade-off. This work proposes a sensitivity-aware privacy budget selection framework for EEG-based analysis under LDP, leveraging multi-criteria decision making to jointly consider privacy level, data accuracy, computational cost, and data sensitivity. Unlike existing approaches that rely on a single global privacy budget, the proposed framework performs participant-specific privacy budget selection, accounting for differences in data sensitivity across participating data sources. Privacy budgets are adaptively selected using TOPSIS and Fuzzy TOPSIS, enabling sensitivity-aware calibration without modifying the underlying LDP mechanism. Experiments conducted on the University of Bonn EEG dataset employ time-domain, frequency-domain, nonlinear, and wavelet features, with eight machine learning classifiers evaluated on privacy-preserving data against a noise-free baseline. The results show that Fuzzy TOPSIS consistently selects smaller privacy budgets, achieving stronger privacy guarantees while maintaining a favorable balance between privacy protection and classification performance. Performance remains stable across classifiers, and the additional computational overhead is modest. Overall, the findings demonstrate that sensitivity-aware privacy budget selection enables effective privacy–utility trade-offs for EEG analysis, offering a scalable and ethically grounded solution for privacy-preserving biomedical signal processing in decentralized healthcare systems.
The implementation of IDSs for intrusion detection based on machine learning has been the subject of extensive research to address the growing threats posed by network intrusions. Nonetheless, there are several problems that must be resolved. Poor detection rates are generated for unknown threats because they are challenging to handle when they do not occur in the training set. IDSs can have a high risk of false positives. Since different models learn data features from different angles, our work introduces a hybrid intrusion detection system (IDS) that combines Random Forest (RF) and Autoencoder (AE). This study implements the combination of RF and AE intrusion detection systems. The hybrid model has two distinct operational phases. We employ the RF classifier's probability output in the first phase for analysis. The system determines, in its first phase, whether the examined sample constitutes an attack. The probability output becomes an essential element for identification. To identify unknown attacks, adding a second Autoencoder helps reduce false positives in the second stage. We specifically remove some samples from one attack class from the training set to simulate an unidentified attack in trials. Our proposed method shows a high detection rate when compared to other baselines. Furthermore, an AE detection module integrated into the system helps decrease the false detection rate. replace word moreover.
The evolution of malware continues to provide a challenge to existing detection methods, particularly those based on static analysis and signature-based heuristics. This paper investigates a memory-forensics-driven methodology for malware detection that includes feature refinement, model explainability, efficiency analysis, adversarial robustness testing, and family-disjoint validation. Using the CIC MalMem-2022 dataset, we examined 55 memory-resident features and reduced them to 13 using mutual information and SHapley Additive exPlanations-guided feature refinement (SHAP-GFR). We tested five classifiers: RF, XGBoost, (RF), eXtreme Gradient Boosting (XGBoost), multilayer perceptron (MLP), one-dimensional CNN (1D CNN), and a CNN-LSTM hybrid using a 70
The rapid growth of data-driven applications has produced massive high-dimensional tabular data horizontally partitioned across independent institutions. Under strict privacy regulations and limited trust in third parties, jointly publishing such data while preserving both individual privacy and analytical utility remains an open challenge. Existing differentially private generative adversarial networks for multi-party publishing typically aggregate discriminator feedback with uniform weights and apply static gradient perturbation, leading to imbalanced fitting under heterogeneous distributions and excessive utility loss under tight privacy budgets. This paper proposes D-PGAN, a differentially private generative adversarial framework for multi-party publishing of high-dimensional data. D-PGAN adopts a single central generator with per-party local discriminators, keeping raw data within each party. A global attention mechanism adaptively weights each party’s feedback so the generator focuses on harder-to-fit distributions. Via chain-rule decomposition of the generator gradient, only the discriminator-side sensitive factor is perturbed; the discriminator is then updated by a differentially private Adam optimizer with density-clustering-based adaptive clipping and a square-root decaying noise schedule. At publishing, a Bayesian-posterior cooperative filter rejects samples on which the discriminators disagree, improving utility at no additional privacy cost. Theoretical analysis shows that D-PGAN satisfies (ϵ ,δ ) -differential privacy with respect to each party’s local data. Experiments on the Adult, TPC-E, KDD99, and NLTCS datasets confirm its superior statistical similarity, downstream accuracy and cross-party fairness, and its robustness under tight privacy budgets and larger numbers of parties.
