-
Salivary fingerprinting and neural network identified high-risk periodontitis in diabetes
Lightweight neural networks analyze salivary metabolics via mass spectrometry to screen for periodontitis and diabetes, achieving 91.9% accuracy with minimal computational resources for clinical.
-
FuXi-Air forecasts six pollutants with multimodal data
FuXi-Air combines meteorological, emission, and observational data for high-precision air quality forecasting at scale, generating 72-hour predictions across multiple sites in seconds.
-
Fused deep learning classified early enamel caries with high accuracy
Deep learning framework with quantum-inspired feature fusion achieves 99.33% accuracy for automated enamel caries classification in intraoral photographs with visual explainability.
-
Neural network predicts shifts in extreme weather frequency
Neural networks leverage climate model data to predict how extreme rainfall, hail, and winds will shift geographically as climate changes, accounting for terrain effects.
-
Geospatial foundation models improved tree species mapping accuracy
Foundation models outperform conventional satellite methods for tree species classification in mountain forests, achieving high accuracy with minimal training data but requiring nonlinear classifiers.
-
Acceptance of generative AI and AI literacy vary across teacher candidates
Mixed methods study of 723 prospective teachers finds GenAI acceptance and AI literacy vary by discipline, grade, tool use, and proficiency, with formal training increasing literacy.
-
RNN-based distortion models improved CAT bond pricing
Catastrophe bond pricing framework combining distortion operator theory with recurrent neural networks, capturing discontinuous repricing and tail-risk compensation.
-
Lower bit depth reduced speaker recognition accuracy
Quantization of neural network output tensors reduces storage for speaker recognition databases. Study evaluates bit depth reduction impacts on recognition accuracy across three architectures.
-
Adaptive music generation improved emotional matching
Emotion-Conditioned Deep Reinforcement Learning framework for adaptive music generation. Achieves 98% emotion mapping accuracy with 280ms real-time responsiveness, enabling dynamic musical.
-
Vision–language model improved pediatric dental disease classification
Deep learning vision-language model for diagnosing pediatric dental diseases in panoramic radiographs, combining visual and textual information with 90% accuracy for caries and periapical.
-
Graph methods may improve major depressive disorder diagnosis
Graph neural networks with augmented brain signals improve MDD diagnosis through gender-specific and stage-wise analysis, enabling personalized therapeutic strategies.
