Adaptive Layerwise Perturbation for Stable LLM RL Training
Discover how Adaptive Layerwise Perturbation (ALP) improves off-policy corrections, stabilizing training and boosting performance in LLM reinforcement lear...
Fine-Grained Solar Irradiance Forecasting with Baguan-Solar
Discover Baguan-Solar, a cutting-edge model combining weather data and satellite imagery for precise 24-hour solar irradiance forecasts at kilometer resolu...
Volumetric Motion Fields for Radar Precipitation Nowcasting
Explore physics-informed deep learning to improve radar-based precipitation nowcasting using volumetric motion fields for enhanced weather forecasting accu...
SciMDR Dataset Boosts Scientific Multimodal Reasoning AI
Discover SciMDR, a large-scale dataset enhancing AI's ability to reason across scientific multimodal documents with 300K QA pairs and expert benchmarks.
Enhancing Harmonic Loss with Non-Euclidean Distance Metrics
Explore how non-Euclidean distance layers improve harmonic loss for better model accuracy, interpretability, and sustainability in deep learning.