2787 papers
The paper proposes TDCO-Net, a Triple-Dynamic Collaborative Optimization Network that enhances multimodal sentiment analysis by integrating dynamic modal auxiliary learning, KAN-based visual-acoustic polarity calibration, and confidence-driven sample selection to effectively address modality reliability variations and boundary sample noise.
This paper proposes CAMF-Zero, a contradiction-aware reasoning framework that leverages a novel Consistency-Aware Contradiction Score (CACS) to detect zero-day threats in federated multi-cloud environments by explicitly evaluating semantic disagreements across heterogeneous telemetry sources, thereby significantly improving detection recall and reducing false positives compared to existing methods.
This paper proposes a robust and explainable multimodal fusion framework that integrates visual and textual features with a stacked ensemble learning approach to achieve state-of-the-art accuracy (98.41%) in emotion recognition, outperforming unimodal methods while ensuring interpretability through SHAP and LIME.
This paper proposes a robust GDD-TL method, which enhances transfer learning with specific layers to achieve a 94.33% accuracy in classifying butterfly species, outperforming standard CNN architectures and demonstrating strong generalization across diverse datasets.
This paper demonstrates that a transparent, domain-stripped language model fingerprint derived solely from abstracts can significantly outperform citation-based metrics in predicting research replicability by isolating a distinct "verifiability" axis, thereby explaining why citations fail as a proxy for scientific robustness.
This study demonstrates that using data science and artificial intelligence to systematically analyze publication metadata and construct graph-based models is a scalable, efficient, and unbiased alternative to manual methods for building clinical and academic health research networks focused on alcohol, addiction, and mental health in Northern England.
This multicenter study presents a novel framework combining a Gabor-prior-guided diffusion model for synthetic data augmentation and a state-space-based segmentation network, which collectively improved pericardial effusion delineation accuracy and demonstrated robust generalizability across internal and external echocardiographic datasets.
This paper introduces AIDA-AND, a cooperative multi-agent framework that integrates semantic representations with interpretable metadata evidence through specialized agents to achieve competitive, modular, and efficient author name disambiguation, significantly reducing execution time while improving performance on benchmark datasets.
This paper proposes SETU, a culturally responsive and affect-aware conceptual framework that bridges India's multilingual classrooms and formal curricula by integrating six components to augment teachers and align AI-assisted learning with the National Education Policy 2020.
This paper audits a LoRA rank comparison study to demonstrate that an inverse-rank compute normalization rule artificially penalizes larger ranks through extreme undertraining, revealing that observed performance rankings are highly dependent on the specific budget allocation rather than inherent model capability.