With the growing adoption of hybrid cloud storage, the identification and protection of sensitive information within large-scale unstructured data has become increasingly challenging. Traditional rule-based and machine learning approaches have limitations in context-aware sensitive data classification and large-scale processing. In this work, a novel framework named LLM-SPSS, implementing a secure and confidential storage layout for distributed hybrid clouds through a fine-tuned XLM-R Base model and multi-dimensional data partitioning, is proposed. First, a fine-tuned XLM-R Base model with adaptive prompt tuning is employed to enable context-aware sensitive data classification and improve detection accuracy. In addition, MapReduce-based distributed processing allows the framework to scale efficiently to large datasets, thus enhancing computational efficiency. Furthermore, a multi-dimensional cloud partitioning scheme provides secure and fine-grained storage isolation within hybrid cloud environments. Experimental results demonstrate that LLM-SPSS achieves an F1-score of 99.66% and yields a 6.3× speed-up over the non-distributed baseline, outperforming traditional rule-based (F1 68.27%), conventional machine learning (SVM F1 98.32%, Random Forest F1 95.79%), and other LLM-based approaches (DePrompt F1 95.95%) and effectively balancing high accuracy with computational efficiency.
In this work, we concentrate on identifying bound entangled states with positive partial transposition in tripartite system. Since directly extending PPT-BE states from bipartite to tripartite systems is nontrivial, we adopt unextendible product bases (UPBs) as a tool to generate tripartite PPT-BE states. First, a set of UPB with cardinalities ${{d}^{3}}-4d+3$ is presented to characterize the extremal states of PPT set in the ${{C}^{d}}\otimes {{C}^{d}}\otimes {{C}^{d}}$ system. Using this basis, the missing states are introduced and employed to identify a $2\left( d-1 \right)$-parameter family of PPT entangled edge states. Additionally, a generation circuit is provided which transforms the states in Z-basis to the missing states in three-qudit composite system. This circuit intuitively reflects the structure of the missing states and prepares for the generation of PPT entangled states. Our work demonstrates a different construction method of PPT entangled edge states in higher-dimensional multipartite system, providing a general insight into the relationship between nonlocality and entanglement in such system.
We present a trusted-dealer Learning with Errors (LWE)-based threshold key-management framework that separates the public-key layer from threshold-share maintenance. The coordinator generates one LWE public key and distributes the same secret vector through coordinate-wise Shamir sharing. Dynamic membership is restricted to enrollment: a new participant obtains an evaluation of the existing sharing polynomial, so enrollment does not rotate the secret or change the LWE key pair. Share refresh uses participant-generated degree- vector polynomials with zero constant term. Their evaluations are exchanged over authenticated private channels and added locally, preserving the same secret without interpolation or accumulation of LWE error. This paper formalizes exact threshold recovery, single-epoch share privacy, and cross-epoch share-snapshot privacy under a restricted post-refresh observation model with hidden refresh state and effective erasure. It also introduces an explicit epoch-consistency rule so that recovery never mixes shares from different refresh epochs and derives communication and computation costs separately for setup, enrollment, refresh, and recovery. Hash-based commitments and authenticated private channels are treated as integrity and delivery mechanisms rather than as substitutes for maliciously secure verifiable secret sharing (VSS). The post-quantum claim is conditional on concrete validation of the selected LWE parameter tuple. Dealerless distributed key generation (DKG), coordinator-compromise proactive security, participant departure, threshold changes, and a complete quantum random oracle model (QROM) reduction remain outside the present security claims.
Large-scale and increasingly sophisticated DDoS attacks expose the limitations of traditional rule-based and architecture-dependent defense methods in accuracy, timeliness, and deployment flexibility. This paper proposes an end-to-end DDoS detection and traceback system for general IP networks, combining network flow analysis with multi-task deep learning. The system employs a hybrid framework that integrates a Random Forest-based feature importance module with a hard-parameter-sharing encoder-decoder architecture using three-layer BiLSTM networks for joint attack classification and feature reconstruction. A context-aware adaptive inference engine further enables dynamic switching between lightweight ensemble models and deep neural networks based on real-time resource conditions. Experiments on the CIC-DDoS2019 dataset show that the system achieves 81.2% detection accuracy, 45 ms average inference latency, and 3.4 s per-IP traceback time. The results demonstrate that the integration of hybrid modeling, adaptive scheduling, and parallel traceback provides a practical and scalable approach for next-generation adaptive DDoS defense.