2615 papers
This study proposes a multimodal deep learning framework that fuses audio and video features using decision-level stacking to achieve a 98% accuracy in human fall detection, significantly outperforming single-modality models and other fusion strategies.
This paper presents a robust, semantically aware multimodal phishing detection framework that fuses URL, text, and visual features with calibrated confidence and adversarial training to significantly outperform unimodal baselines while maintaining high accuracy and reliability under deceptive attacks.
This article presents an artefact-based case study of AgentGuard AI v2.4.0, a research prototype that operationalizes runtime risk detection and control mechanisms for AI agents, while highlighting its utility as an inspectable governance instrument and identifying specific reproducibility defects that preclude claims of production effectiveness.
This paper presents a brain-like self-organizing hardware architecture that utilizes electrochemical mechanisms to enable in-materia learning and memory through the co-evolution of structural dynamics and functional connectivity, thereby achieving embodied intelligence via adaptive dendritic memristive networks.
This paper presents a framework for EEG-based imagined handwriting recognition that combines a frozen cross-subject neural encoder with a lightweight target-specific classifier and a confidence-aware trie decoder to achieve rapid personalization, high word-reconstruction accuracy, and low-latency energy-efficient performance on edge devices.
This reproducible multi-dataset evaluation demonstrates that adaptive feature-group masking does not significantly outperform simpler, non-adaptive corruption policies in improving robustness against feature-set shift, suggesting that matched controls and dataset-level inference are more critical than complex adaptation mechanisms for this task.
This paper proposes a closed-loop, data-driven framework that integrates process mining, complex network analysis, explainable machine learning, and digital twin simulation to automate Value Stream Mapping and quantitatively optimize Future-State designs, demonstrating significant reductions in lead time, work-in-process, and defects across real-world manufacturing scenarios.
This paper proposes a robust framework for discovering governing equations of fractional-order dynamical systems from noisy data by integrating a Deep Operator Network (DeepONet) for accurate fractional derivative estimation with ensemble Sparse Identification of Nonlinear Dynamics (SINDy) for sparse regression, demonstrating superior accuracy and generalization compared to classical finite difference methods and standard neural networks.
This study demonstrates that using a small validation budget to select a compact text representation (specifically a TF-IDF and SVD combination) for e-commerce recommendation tasks can reduce computational costs by up to 90% while maintaining test performance comparable to full-budget selection, provided the candidate pool is appropriately constrained.
This study demonstrates that enforcing explicit evidence-admission requirements in clinical question answering significantly alters route rankings and latency metrics while reducing material evidence-validity errors, highlighting the need for standardized admission rules in clinical latency benchmarks.