863 papers
This paper proposes an intelligent, real-time drone detection and early warning framework for critical infrastructure surveillance using the YOLO26 deep learning model, which is trained on a custom multi-scenario database and evaluated against baseline models to demonstrate superior accuracy and speed in identifying small, distant, and challenging drone targets.
This paper demonstrates that Recursive Indexing (RIX) analysis of latent transport topology can detect structural regime transitions in world models that precede behavioral failure, revealing that representations can remain pointwise stable while their underlying organizational structure collapses under stress.
Tamarin (HSQ) introduces a reversible, three-tier hierarchical KV-cache compression scheme that routes tokens into a focal tier, learned summary vectors, and a CPU-based archive to achieve 12–28× GPU memory reduction for long-context LLM inference while maintaining near-baseline perplexity and retrieval accuracy.
This paper proposes a graded-autonomy framework, adapted from the automotive J3016 standard, to govern agentic AI in healthcare by classifying systems across three domains and six levels to separate capability from authorization, thereby addressing safety gaps in current evaluation metrics through a worked example of autonomous sepsis management.
This paper proposes an adaptive-hazard Bayesian online change-point detection method for text streams that dynamically adjusts reset probabilities based on lexical distribution drift via a Dirichlet-multinomial formulation, demonstrating improved detection performance and reduced delay particularly in scenarios involving gradual or weak lexical changes.
This paper demonstrates that an Isolation Forest algorithm combined with Mutual Information-based feature selection outperforms Autoencoder and Ensemble methods in large-scale IoT intrusion detection by achieving superior accuracy, F1-score, and AUC while maintaining lower computational complexity on the NF-ToN-IoT-v3 dataset.
This paper introduces PromptShield, an interpretable multi-agent framework that combines transformer-based semantic analysis with XAI techniques like SHAP and LIME to effectively detect and explain prompt injection attacks in Large Language Models.
This paper proposes a temporal multi-signal fusion approach that treats hallucination as a sequence-labeling problem using a BiGRU to combine text statistics, NLI entailment, and language model surprisal, achieving a significant AUC improvement of 0.840 on RAGTruth by leveraging temporal context rather than model capacity, while maintaining effectiveness on closed-source and unseen models.
This paper proposes a risk-oriented verification and validation framework that transforms hallucination detection in public-sector LLMs from a purely technical challenge into a governance issue by introducing a composite risk index to guide operational decisions like deferral or human escalation, thereby ensuring legal certainty and administrative accountability.
This paper introduces a causal, token-level survival analysis framework that successfully forecasts the onset of hallucinations in large language models before they occur, achieving a 0.777 AUROC by detecting text-novelty drift and providing warnings approximately 11 tokens earlier than traditional post-hoc detectors.