2786 papers
The paper proposes CTDG-IDS, a continuous-time dynamic graph representation learning framework that integrates memory updating, hypergraph construction, and a dynamic graph attention mechanism to effectively detect network intrusions by capturing both evolving topological structures and temporal interactions.
This study evaluates GitHub's potential as an integrated research infrastructure, finding that while its native capabilities effectively support execution and output phases, they offer only partial traceability for upstream research activities like question formulation and require complementary tools for full lifecycle management.
CleanScore is a black-box auditing framework that employs negative controls and sensitivity bounds to distinguish genuine model skill from prior data exposure, revealing that standard paraphrase-based audits significantly underestimate the impact of training data contamination while demonstrating that surface-form inflation in public benchmarks is minimal.
This paper proposes a Neural Screener–Reasoner framework that combines a parser-free, linear self-attention detector with a retrieval-augmented LLM to achieve high-accuracy log anomaly detection and generate verifiable, evidence-grounded explanations while managing inference costs.
This paper proposes Boundary-First Segmentation, a training-free paradigm that constructs closed query regions before identity arbitration using Unified Boundary Evidence Transport (UBET), achieving state-of-the-art performance across diverse benchmarks by prioritizing spatial coherence over dense semantic similarity.
This paper proposes a quantum-resilient cryptographic framework called the Fractal Spectral Wave Filter, which achieves efficient O(N log N) processing and high entropy by mapping 2D data via Morton Z-order curves, applying Number Theoretic Transforms over finite rings, and leveraging unmitigated aliasing collisions as a diffusion mechanism.
This paper introduces Mastyf Guard, a deterministic-first security architecture that treats LLM agents as untrusted principals and enforces conditional non-escalation invariants via an external reference monitor, achieving near-perfect attack detection with zero false positives in empirical evaluations against indirect prompt injection and exfiltration threats.
This paper presents and validates a secure, offline wearable vision architecture integrating an ESP32-CAM and Raspberry Pi that enforces authenticated encryption, replay resistance, and fail-closed startup through reproducible software contracts and rigorous CI-based testing.
This paper presents the design and evaluation of a low-cost, Raspberry Pi-based robotic face that utilizes OpenCV for real-time visual tracking and servo-controlled actuation to mimic human facial expressions and head movements, achieving 92.4% detection accuracy with minimal latency while addressing security considerations for intelligent interactive systems.
This paper proposes an improved Deep Q-Learning algorithm for UAV fire detection in educational experimental zones that integrates multi-objective reward optimization, diagonal motion incentives, and A*-inspired constraints to significantly enhance path efficiency, reduce risk exposure, and improve mission success rates compared to conventional methods.