We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
DeepMind Blog·Nov 17·5 min read
WeatherNext 2: Our most advanced weather forecasting model
TL;DR
Weather forecasting has traditionally struggled with accuracy and speed, leading to less reliable predictions. DeepMind's WeatherNext 2 model uses advanced AI techniques to generate hundreds of weather scenarios in under a minute.
✦ Why It Matters
Engineers and researchers can leverage WeatherNext 2 for more accurate weather data in their applications.
Key Takeaways
- Access WeatherNext 2 data via Earth Engine and BigQuery.
- Utilize the early access program on Google Cloud's Vertex AI.
- Incorporate advanced weather predictions into your applications.
- Leverage hundreds of scenarios for better planning and decision-making.
How It Works
WeatherNext 2 employs a Functional Generative Network (FGN) that introduces variability into the model architecture. This allows it to generate coherent weather predictions while ensuring they remain physically realistic.
By training on individual weather elements, or 'marginals', the model learns to forecast complex interdependencies, or 'joints', which are essential for accurate regional weather predictions.