2890 papers
This paper presents a low-resource hybrid detection pipeline that combines motion-based preprocessing with YOLOv8-nano to efficiently identify Red-billed Quelea swarms on CPU-only hardware, achieving high accuracy while significantly reducing computational load during static scenes.
This paper introduces a quantum-inspired tensor network framework using Matrix Product States and AlphaEarth embeddings for wildfire susceptibility mapping in the Gargano region, which not only achieves competitive classification accuracy but also reveals distinct grokking dynamics and provides a physically grounded, level-resolved analysis of class separability through quantum mixedness diagnostics.
This paper proposes HCSC-Net, a hierarchical contextual semantic calibration framework that enhances CLIP-based weakly supervised semantic segmentation by integrating Group-wise Structural Calibration and Residual Contextual Semantic Calibration modules to resolve issues of structural inconsistency and background interference, achieving state-of-the-art performance on PASCAL VOC 2012 and MS COCO 2014 benchmarks.
This randomized controlled study reveals that large language models are significantly more susceptible than human clinicians to contextual cues that induce incorrect diagnostic choices, highlighting a critical safety vulnerability for clinical AI systems that can be mitigated through specific prompting strategies.
This paper presents an explainable PCB defect classification framework that combines TRIBRID-optimized hyperparameter tuning, transfer learning, and MobileNetV3 to achieve perfect accuracy and interpretability while maintaining the computational efficiency required for intelligent Automated Optical Inspection systems.
This paper presents a hardware-independent post-quantum attestation mechanism for legacy industrial devices that leverages the hibernation-resume cycle on commodity x86 platforms to capture and sign runtime state using CRYSTALS-Dilithium, thereby enabling trustworthy verification without dedicated security hardware like TPMs.
This paper proposes a Hybrid Multi-Expert Deep Learning Framework that utilizes an intelligent ResNet-50 routing mechanism to dynamically select specialized detection models (YOLOv8 and SwinIR-enhanced Mask R-CNN) for different maritime environments, achieving high accuracy in ship detection and successfully identifying dark vessels in real-world scenarios like the Singapore Strait.
This paper demonstrates that pretrained diffusion models implicitly encode population-level anatomical atlases within their learned dynamics, enabling the direct recovery of sharp, age-conditioned, and registration-competitive brain templates through deterministic inference without requiring explicit registration or atlas-specific training objectives.
This paper introduces BQEB ForecastBench, an open benchmarking framework utilizing the BQEB-Data v1 dataset to standardize the evaluation of AI models for smart-grid forecasting, demonstrating through baseline experiments that Linear Regression outperforms other models in predicting electricity load and prices while highlighting the inherent complexity of price forecasting.
This study proposes a dual-task deep learning framework based on LSTM that simultaneously predicts customer repurchase timing and next-product selection, demonstrating superior performance over existing models on large-scale e-commerce data and offering a practical AI agent system for automated marketing optimization.