2786 papers
RheumBench is the first specialist-validated, rubric-based benchmark for rheumatology that evaluates ten LLMs and RAG systems, revealing that current models exhibit substantial performance gaps, systematic overconfidence, and potentially severe omission errors, thereby underscoring the critical need for human oversight in clinical decision support.
This systematic literature review analyzes the paradigm shift in object detection from traditional convolutional architectures to Transformer-based and few-shot learning models, highlighting their benefits in simplifying detection pipelines and improving data efficiency while addressing persistent challenges like computational cost and small object detection through architectural innovations and advanced pre-training strategies.
This preregistered study finds that provenance metadata in agent skills does not reliably improve task utility beyond matched ordinary guidance, suggesting its value is context-dependent and primarily realized only when it alters executable actions rather than through bundled skill scores alone.
Policy-TD enhances the reliability of frozen small language models by distilling teacher-audited trace states into a runtime controller that intervenes to fix answer-commitment failures, significantly improving internal and stress-test accuracy without degrading performance or creating new capabilities.
This paper introduces TOAIAF, a runtime AI assurance control plane featuring a three-stage decision procedure and seven governance components that significantly outperforms baseline systems in detecting safety violations, prompt injections, and privacy risks for on-device agentic AI, while transparently reporting experimental limitations and mapping its framework to major international AI safety standards.
This study proposes a UAV-based deep learning system utilizing a fine-tuned Mask R-CNN algorithm to detect drifting victims in flash floods and estimate their trajectories using GPS metadata, achieving over 90% detection confidence to enhance Search and Rescue operations.
This paper introduces SpikeGate, an ultra-low-power, event-driven Spiking Neural Network gatekeeper that significantly reduces energy consumption for continuous on-device Small Language Model inference by suppressing routine background signals to duty-cycle transformer execution only during detected anomalies.
This study demonstrates that incorporating statistical relationships into quantum feature encoding for Quantum Support Vector Machines can influence binary classification performance, but emphasizes that optimal strategies require balancing predictive accuracy with circuit complexity rather than simply increasing entanglement.
This study developed and evaluated a transparent, context-aware retrieval system for mental health and substance use services using the HealthLink BC directory, demonstrating that combining semantic search with explicit context weighting improves ranking performance over lexical baselines while highlighting the need for further validation with real-world data and human relevance judgments before clinical deployment.
This paper proposes ASWA, a communication-efficient Federated Learning algorithm that enhances accuracy and reduces communication rounds under Non-IID data by adaptively weighting client updates based on data size and update similarity while prioritizing clients with higher loss, all without requiring additional data transmission.