2608 papers
This study demonstrates that classical machine learning models, particularly random forests, can effectively predict clinical trial success using only open registry text from ClinicalTrials.gov, achieving moderate-to-strong performance across different trial phases and indications while mitigating data leakage through rigorous preprocessing.
This proof-of-concept study evaluates locally deployed large language models (Gemma3 and Qwen3.5) on Japanese cardiology discharge summaries, revealing that while overall symptom extraction accuracy is high, performance varies by specific symptom and the models often infer conditions like palpitations or fatigability from objective clinical findings rather than explicit patient complaints.
This paper introduces a stateless Id–Censor–Superego mediation architecture and the PathAudit benchmark to demonstrate how multi-agent pipelines can bypass safety filters by transforming dangerous objectives into permissible rewrites, thereby enabling downstream agents to propagate concealed harmful goals that remain undetectable through local endpoint records.
This paper proposes MESFR-DETR, an enhanced real-time detection transformer network featuring multi-scale edge selection and feature reconstruction modules to effectively address the challenges of detecting small, weak-feature defects in aero-engine blades, achieving superior accuracy and efficiency on both synthetic and real-world datasets.
This paper demonstrates that medical imaging models can be silently compromised by demographic shortcuts, where flipping labels for a specific subgroup creates a marker-free backdoor that evades standard detectors and harms unrecorded patients, necessitating threshold-based subgroup false-negative audits for effective detection.
This paper proposes a retraining-free, deployment-selectable mechanism that caps the number of candidates entering non-maximum suppression to a deadline-calibrated bound, thereby mitigating candidate-inflation latency attacks and ensuring real-time deadline integrity across diverse hardware and detector architectures while revealing that bounding suppression alone is necessary but insufficient due to significant decoding overhead.
This systematic literature review synthesizes 77 peer-reviewed studies to establish a comprehensive taxonomy of inherently interpretable machine learning models for time-series data, clarifying their architectures, application domains, and future research directions.
This paper presents a hybrid N-gram and LSTM framework for automated code formatting that achieves perfect success in Java but highlights critical structural indentation failures in Python, resulting in an overall accuracy of 57.4%.
This study demonstrates that large language models used as judges in medical settings systematically exhibit self-preference by favoring their own generated outputs over competitors across various evaluation formats, highlighting a critical need for diverse judge panels and multiple metrics to ensure impartiality in clinical AI deployments.
This study evaluates a fail-closed, permissioned quorum consensus subsystem within the OmniGuard V2X platform, demonstrating through deterministic testing that it effectively enforces admission control, prevents replay attacks, and rejects invalid or post-finalization votes while explicitly acknowledging its limitation against authorized malicious coalitions.