833 papers
This study utilizes multimodal wearable data and machine learning to demonstrate significant stress differences between English Learners and English Speakers during digital assessments, achieving high detection accuracy and identifying key predictive features to inform the design of more supportive, human-centered educational technologies.
This paper proposes the Tangent Subspace Boundary Attack (TSBA), a query-efficient decision-based black-box adversarial attack that improves upon existing methods by constraining perturbation updates within a low-dimensional tangent subspace of the decision boundary to stabilize the search process and significantly reduce query complexity while maintaining competitive distortion levels.
This study demonstrates that while Kolmogorov–Arnold Networks (KANs) inherently mitigate spectral bias better than standard MLPs in nonlinear structural dynamics, the application of Fourier feature encoding is the dominant factor for accuracy, making Fourier-enhanced MLPs superior to both plain KANs and Fourier-enhanced KANs, which suffer performance degradation due to gradient oscillations incompatible with physics-informed optimization.
This paper introduces IRA, a hybrid conversational framework that combines an interpretable MLP intent classifier trained on a code-mixed corpus with a generative fallback mode to achieve high accuracy and transparency in emotion-aware dialogue for mental-health and support applications, noting that duplication-based oversampling does not introduce lexical variation.
This paper addresses the challenge of providing individualized feedback in large-scale university badminton classes by introducing a new benchmark dataset of 7,200 annotated video clips and a baseline model that automatically evaluates student movement quality against professional demonstrations, achieving promising performance metrics for intelligent physical education assistance.
This paper proposes that effective reasoning in large language models emerges not merely from low-dimensional inference manifolds, but from a specific constrained structural regime balancing expressivity, compression, and information preservation, enabling a new label-free diagnostic framework based on internal geometric dynamics.
This paper introduces Aux-AI, a federated independent oversight framework featuring a decoupled Trust-Decision Core that ensures trust-aware industrial autonomy by effectively detecting hardware-induced failures and out-of-distribution shifts with superior accuracy, real-time throughput, and energy efficiency compared to existing safety monitors.
TopoNet is a topology-guided dual-branch framework that enhances volumetric medical image classification by integrating spatially registered topological features derived from persistent homology with appearance-based cues through cross-branch attention, thereby improving accuracy, robustness, and interpretability.
This paper identifies and analyzes CVE-2026-0628, a critical privilege escalation vulnerability in Google Chrome's Gemini Live AI side panel that allows extensions with basic permissions to access sensitive browser-level resources, revealing that 73.9% of analyzed extensions possess the necessary permissions to exploit this security gap.
This study demonstrates that using large language models as editorial assistants to refine physicians' written responses, rather than as autonomous generators, significantly enhances perceived empathy while preserving factual accuracy, supported by novel quantitative metrics for evaluating both emotional tone and medical precision.