←── back to feed
/topics/time-series-and-forecasting-neural-networks
Time series and forecasting neural networks
2 items●1 sources●updated 27d ago●trend 0
Two arXiv papers advance neural network forecasting: one proposes QHAdamW, a modified optimizer combining quasi-hyperbolic and decoupled weight decay techniques for air quality index prediction in the Philippines, while the other introduces PIHIM, a physics-informed hybrid neural model for Arctic sea ice concentration forecasting that integrates physical constraints with data-driven learning.
- QHAdamW optimizer addresses convergence, generalization, and forecasting performance limitations of standard Adam
- Study presents first AQI forecasting model available in the Philippines using enhanced neural networks
- PIHIM integrates physical dependencies explicitly into neural architecture for sea ice concentration modeling
- Physics-informed approach reduces computational complexity versus traditional numerical parameterization methods
- Both papers target environmental forecasting: air quality and polar climate prediction applications