With the proliferation of dynamic network environments such as the Internet of Things (IoT) and Vehicle-to-Everything (V2X) networks, challenges related to data localization and privacy leakage have become increasingly prominent. Efficient and privacy-preserving real-time network attack detection has emerged as a critical issue in the field of cybersecurity. Existing methods struggle to balance handling dynamic topology changes, resisting privacy attacks, and maintaining high detection accuracy. To address this, this paper proposes an attack detection framework that integrates adaptive spatio-temporal graph generation, spatio-temporal Transformer, and privacy-enhanced federated learning. The approach designs a dynamic adjacency matrix optimization mechanism to capture real-time communication relationships, incorporates gradient compression and feature decoupling strategies to effectively mitigate model degradation caused by differential privacy noi se, and introduces a deep reinforcement learning module to dynamically adjust defense strategies based on real-time traffic feedback. This enables efficient collaborative training while ensuring data remains localized. Experimental results demonstrate that the proposed method achieves accuracy rates of 93.5% and 92.0% on the UNSW-NB15 and BoT-IoT datasets, respectively. Specifically, the proposed method achieves a 3.5 percentage point improvement over the strongest privacy-preserving baseline (DP-GNN) on the BoT-IoT dataset. Moreover, under the same privacy budget, it significantly reduces the success rate of membership inference attacks. This work provides novel insights into distributed secure learning in high-privacy-risk dynamic environments, with its low-latency and high-robustness characteristics supporting real-time threat perception and privacy-secure responses on edge devices.
The Number Theoretic Transform (NTT), a finite-field variant of FFT, is critical in cryptography and digital signal processing, but its efficiency on FT-M7032 Digital Signal Processors(DSPs) remains suboptimal due to memory bottlenecks and architectural constraints. This paper proposes a tailored optimization framework: a four-step strategy decomposes large 1D NTT into 2D transforms, outperforming traditional six-step methods by reducing memory access overhead; a VLIW/SIMD-aware microkernel with loop unrolling and register reuse boosts computation; vectorized Montgomery modular multiplication supports 16 parallel streams to address the lack of division instructions; and double buffering with pipelining overlaps computation and data transfer. Experiments show a 90.6 & times; speedup over baselines for large-scale NTTs, with ablation studies validating each optimization. This work offers insights for NTT acceleration on DSPs and data-intensive algorithm optimization on heterogeneous processors.
Fuzzing has emerged as a vital technique for discovering vulnerabilities in software systems. American Fuzzy Lop (AFL) is a widely used fuzzer known for its effectiveness and efficiency. However, optimizing fuzzing strategies to enhance coverage and bug discovery remains a significant challenge. This paper introduces an innovative approach by integrating an Adaptive Reward Mechanism within the Proximal Policy Optimization (PPO) framework to improve AFL fuzzing performance. Our method dynamically adjusts rewards based on the fuzzing feedback, guiding the learning agent to prioritize paths with higher potential for uncovering vulnerabilities. Experimental evaluations demonstrate a clear trend of improvement in both code coverage and unique crashes found as model iterations increase. The Adaptive PPO method consistently outperforms the traditional AFL and standard PPO techniques across all iterations. By the 100th iteration, Adaptive PPO achieves 90% code coverage and discovers 25 unique crashes, compared to AFL's 60% coverage and 11 crashes, and PPO's 78% coverage and 17 crashes. This data highlights the effectiveness of the adaptive reward mechanism in enhancing exploration and vulnerability detection capabilities, supporting the assertion that Adaptive PPO is a superior approach for fuzz testing. The incremental improvements observed with each iteration underscore the method's potential for sustained performance gains over time, advancing the state-of-the-art in software testing and vulnerability analysis. Additionally, performance overhead analysis shows the method’s practical feasibility." to emphasize overhead validation.
Existing Automatic Exploit Generation (AEG) solutions primarily target Linux and Windows binary files. Currently, only one AEG framework for IoT devices—IoTAEG—has been proposed, which emphasizes bypassing diverse protection mechanisms and relies on real devices for exploit generation. To enhance the efficiency of exploit generation for IoT firmware vulnerabilities, we present Automatic Exploit Generation Based on Shared Keywords (SKAEG) in embedded systems, a novel shared-keyword-based approach. SKAEG automatically detects and analyses vulnerabilities within IoT device firmware and generates exploits subsequently. It first extracts shared keywords from firmware to guide taint analysis for vulnerability detection. To improve efficiency, pruning strategies are applied during taint analysis to eliminate infeasible paths. Subsequently, we introduce two new modules: Exploitability Analysis, which evaluates whether identified vulnerabilities can be exploited, and Frontend File Parsing, which constructs request packets based on shared keywords. This design enables SKAEG to integrate vulnerability analysis with concrete exploit generation, thereby facilitating the automated construction of complete exploits. Experimental results demonstrate that SKAEG successfully detects and exploits vulnerabilities across 10 IoT devices from 4 manufacturers, covering 8 CVE-listed vulnerabilities. Additionally, the exploitability analysis accurately identifies 51 non-exploitable vulnerabilities. Through optimized symbolic execution, SKAEG addresses solver crash issues and achieves superior exploit generation efficiency compared to IoTAEG.
Veritas Gateway is a zero-knowledge verification layer that acts as an intermediary between AI agents and diverse data stores through the Model Context Protocol (MCP). Database snapshots are committed to with Poseidon based Merkle commitments and query answers are returned with succinct SNARK proofs that can be locally verified without satisfying predicates on raw values or trusted hardware. The prototype implements this stack across relational, NoSQL, time-series and vector databases through generic adapters and is empirically evaluated along four dimensions: scalability across MySQL as snapshot size and Merkle height varies, a hash-primitive ablation study with Poseidon, MiMC-7 and SHA-256, a proof-system ablation across Groth16, PLONK and STARKs on a common circuit, and horizontal portability across SQLite, PostgreSQL, MongoDB,InfluxDB and Qdrant under a fixed workload. Measurements include build time, proving time, query time, query assembly time, query verification, memory footprint, and derived throughputs. It is found that build time increases predictably with snapshot size and height, proving cost is driven by path depth and is consistent across iterations, query assembly time is consistently sub-second, and verification completes in approximate to 8-12 ms across all settings. Poseidon incurs the lowest in-circuit latency and variance, with SHA-256 only competitive for out-of-circuit pre-hashing. Groth16 results in the shortest verified-latency path, PLONK results in a reduction of proving time, and STARKs result in transparent setup at a higher memory cost. Cross-backend variation is dominated by build overhead and the data-path overheads for each backend's cryptographic verification, for which cost is never a bottleneck. These results suggest that Veritas Gateway provides backend-agnostic, verifiable querying with predictable latency, improving trust in the MCP-driven agent workflow while maintaining the benefits of tool-based integration and facilitating deployment in privacy-sensitive environments.
