4158 papers
This study introduces a live-agent benchmark for healthcare AI that demonstrates while independent authorization layers did not significantly reduce unsafe actions in synthetic care-coordination episodes, they did significantly degrade routine task success, escalation accuracy, and calibration while increasing time and human effort.
This paper proposes AFCC, an adaptive fusion framework that combines local contour descriptors with hierarchical convolutional features and reliability-guided weighting to achieve high-accuracy classification of weld formations on curved workpieces, significantly outperforming baseline deep learning methods in an industrial setting.
This paper proposes a novel deterministic algorithm that solves the NP-hard Travelling Salesman Problem in strict O(N³) polynomial time by mapping discrete permutation spaces onto a continuous probability-flow tensor manifold and utilizing Maclaurin series expansions to bypass exponential enumeration.
This paper demonstrates that supplying heterogeneous external information cues is a more reliable and effective method for increasing semantic diversity in large language model outputs than relying solely on explicit instructions, highlighting the critical distinction between informational inputs and behavioral guidance in prompt design.
This paper proposes a multimodal digital twin architecture for adaptive art therapy monitoring that integrates facial, speech, physiological, and behavioral data using advanced deep learning models to enable real-time patient state analysis and predictive treatment personalization with high accuracy and low latency.
This paper resolves the open question of the exact zero-error one-way classical communication cost for a specific entanglement-assisted quantum learning task by proving it equals the logarithm of the ternary-cube covering number (or the total domination number of the categorical power ), thereby quantifying the precise classical resources required to match the quantum protocol's perfect success.
This paper introduces InflateBench to demonstrate that while context compaction in LLM agents rarely causes instruction loss, it frequently leads to "constraint inflation" by incorrectly broadening user preferences into strict requirements, a problem that existing fixes for instruction loss often exacerbate.
This paper challenges the assumption that multi-agent systems inherently improve accuracy through collaboration, demonstrating via extensive experiments that homogeneous agentic swarms often exhibit "Architectural Tribalism" where a synthesizer's model family biases the system toward reinforcing errors rather than truth, thereby necessitating a "Heterogeneity Mandate" for resilient AI architectures.
This study demonstrates that while large language models show strong correlations with human raters in measuring complexity, accuracy, and fluency in L2 writing, they exhibit construct-specific biases and systematic underestimations, proving that statistical alignment does not equate to interchangeable measurement and necessitating construct-specific validation and human verification.
This paper proposes a hybrid framework combining Optuna and Q-learning to adaptively optimize hyperparameters in a GRU–HiBERT–MLP model for EEG seizure classification, achieving 93.10% accuracy and enhanced convergence stability while incorporating SHAP and LIME for explainable clinical decision support.