898 papers
This paper proposes a hybrid deep learning framework that integrates transfer learning, CNNs, and attention mechanisms to overcome the limitations of traditional emotion detection systems, achieving superior accuracy, robustness, and cross-domain adaptability for real-world human-computer interaction applications.
This paper proposes a multimodal AI framework that integrates a custom Mask R-CNN for CTPA clot segmentation, a hybrid LSTM-GRU model for ICU time-series analysis, and an artificial neural network for clinical risk assessment, achieving a Dice Score of 0.88 and 91.81% overall accuracy to improve pulmonary embolism diagnosis and patient risk prediction.
The paper proposes UAMG-RAG, a framework that dynamically adjusts retrieval depth and document granularity based on question complexity and employs uncertainty-guided re-retrieval to enhance reasoning performance and reduce hallucinations, achieving significant improvements on HotpotQA and WikiHow benchmarks.
This paper proposes a data-driven, loss-function-free binary classifier that integrates symmetric intuitionistic fuzzy TOPSIS with self-interactive sequential three-way decision to overcome existing limitations by utilizing objective attribute weights, balanced distance measurements, and an internal feedback loop for adaptive thresholding, thereby enhancing classification performance and robustness.
This study presents a preliminary deep learning framework using mel-spectrogram-based convolutional neural networks to automatically distinguish between Maxaatiri and Maay Somali dialects, achieving approximately 81% accuracy on a small, broadcast-derived dataset while highlighting the need for larger, more diverse data to ensure generalizability.
This prospective study demonstrates that the NeutrinoReview tool, utilizing a self-hosted LLaMA 3.1-8B model, can reduce systematic review screening workload by up to 74% without missing any included records, though the authors recommend its use as a conservative adjunct to human screening pending further validation across diverse settings.
This paper proposes FSMF, a reliability-guided RGB-IR object detection network for autonomous driving that employs a reliability-guided modulation module, a complementary information decoupling module, and a scale-aware fusion pyramid to effectively address illumination and texture challenges, achieving state-of-the-art performance on M3FD and FLIR datasets.
The paper introduces AgriMemSynth, a synthetic benchmark framework comprising conversational and multi-hop reasoning datasets to evaluate agricultural AI systems, revealing that retrieval organization strategies are task-dependent and that lexical metrics often underestimate answer correctness.
This paper proposes Meta KG-NCF, a hybrid meta-learning framework that integrates Knowledge Graph embeddings (via TransE) with Neural Collaborative Filtering and a First-Order Multi-Supervisor Association Network to effectively address cold-start problems while achieving superior accuracy and computational efficiency compared to existing baseline methods across multiple large-scale datasets.
The paper proposes Cry2Vec, a self-supervised model leveraging an optimized HuBERT architecture and a large-scale dataset of 3 million unlabeled infant cries to achieve state-of-the-art performance in cry recognition with demonstrated feasibility for edge deployment.