2769 papers
This study demonstrates that certain large language models used as adjudicators exhibit asymmetric, presentation-driven conformity to specific label cues (particularly "FALSE" values) rather than impartially evaluating merits, with susceptibility varying significantly across vendors and being mitigated by instruction tuning.
This paper introduces H-Elena, a compromised coding LLM that demonstrates how trigger-conditioned malicious behaviors can be encoded into model weights and persistently propagated across different architectures through fine-tuning workflows, thereby revealing a critical new supply-chain risk in AI development.
This paper proposes a multi-source data fusion computing model that integrates cable topology, AIS trajectories, and underwater acoustic data via graph convolution and Dempster-Shafer theory to significantly improve submarine cable threat detection accuracy and robustness against spoofing attacks compared to single-source methods.
HUE-OOD is a post-hoc framework that enhances out-of-distribution detection robustness across diverse benchmarks by combining predictive, representational, and evidential uncertainty signals through a parameter-free rank-based fusion strategy, requiring no modifications to pretrained classifiers.
This paper presents a two-stage integer-programming algorithm implemented on a low-code, open-source platform that replaced manual processes for a Tel Aviv short-term rental operator, resulting in a 3.5% revenue increase and a 90% reduction in processing time by optimizing booking allocation rather than pricing.
This paper proposes a pre-deployment validity framework that evaluates network topology through Graph Structural Differentiability and Prominence Isolation to determine whether structural signals are sufficient for effective graph-based intrusion detection, revealing that high structural inequality alone does not guarantee attack-selective performance and that temporal sensitivity must be considered before committing resources.
The paper proposes GapR, a fast and effective multi-boundary ranked-list truncation method that improves multi-hop retrieval performance and downstream answer quality by identifying multiple informative score boundaries rather than relying on a single largest gap, thereby balancing evidence coverage with computational efficiency.
This paper demonstrates through factorial experiments that collective AI outcomes arise from the specific interaction regime where structural and behavioral causes are separable and context-dependent, proving that effective AI assurance cannot target agents or networks in isolation but must instead focus on the dynamic causal structure of their interactions.
This paper establishes a rigorous, leakage-safe benchmarking protocol for credit default prediction that demonstrates gradient boosting ensembles significantly outperform logistic regression and TabNet on imbalanced financial data, while revealing that post-hoc threshold tuning renders synthetic resampling redundant and highlighting the critical need to validate explanation discordance between SHAP and LIME for model risk management.
This paper proposes a methodological trust-boundary security assurance framework that models LLM-integrated and agentic systems to translate evolving semantic threats into verifiable, testable security controls and evidence-based reporting.