4157 papers
This chapter proposes a multimodal deep learning framework that integrates diverse sensing technologies, advanced computer vision models, and autonomous marine robotics to detect, localize, classify, and remove submerged ocean plastic waste, thereby supporting Sustainable Development Goal 14.
The paper proposes DFINE-AGS, a novel Transformer-based detector incorporating global semantic anchoring, content-dependent routing, and phase-preserving convolutions to achieve robust, real-time potato tuber detection in complex field environments with high accuracy and efficient edge deployment.
This paper demonstrates through the WEAVE tool and SHAP-based explainability that machine learning models for detecting security-relevant methods fail on unseen code not due to poor model selection, but because they rely on unstable surface features like identifiers and signatures, identifying feature engineering as the true bottleneck in achieving robust detection.
This study demonstrates that sampling variability across multiple generative inpainting reconstructions serves as a strong predictor of reconstruction error in lacquerware images, enabling the prioritization of outputs for human review despite its inability to guarantee historical authenticity or eliminate errors entirely.
This narrative review critically examines how discrepancies between curated evaluation datasets and real-world operational conditions—particularly regarding low-resource languages like Somali and mobile communication channels—undermine the reliability of current audio deepfake detectors, proposing a comprehensive, evidence-informed evaluation framework to improve transparency and generalization in future research.
The paper introduces CAG-Cap, a retrieval-free framework that enhances image captioning by leveraging self-generated beam hypotheses as auxiliary sequence-level supervision during training, thereby improving model performance on benchmarks like MS COCO and Flickr30k without requiring external retrieval or additional inference-time parameters.
This paper investigates the reliability boundaries of HyDE-style query expansion in multi-hop retrieval, finding that while it significantly improves support retrieval metrics, it frequently fails to translate these gains into higher-quality delivered evidence or improved answer accuracy across various datasets and reader configurations.
This paper introduces VisText-VLM2Plastic, a bilingual English–Bangla benchmark for plastic-waste visual grounding and environmental reasoning, revealing that while state-of-the-art vision-language models achieve high recognition accuracy, they exhibit significant performance gaps between languages and varying reliance on visual evidence across different task types.
This paper introduces CMS2-Fusion, a novel infrared and visible image fusion framework that leverages a cross-modal state space coupling mechanism within Mamba blocks to dynamically interact hidden states during encoding, thereby achieving superior performance and parameter efficiency compared to traditional decoupled approaches.
This paper proposes a secure Smart City framework that prevents gateway-level log falsification by integrating PEKeccak-based on-chain anchoring with advanced cryptographic and machine learning techniques, achieving a 98.91% integrity verification success rate.