Jeff, why is “physics-free” a virtue? Why not also combine with physics, which has powerful tools?
There's great potential in the use of machine learning for a wide variety of weather and climate problems. Here's one example, predicting precipitation at much finer spatial granularity than traditional physics-based models.https://twitter.com/GoogleAI/status/1216795168797409280 …
-
-
-
I suspect that in the end, we'll end with a hybrid system combining the traditional physics-based approaches and this kind of pure ML approach. (cont)
- Još 4 druga odgovora
Novi razgovor -
-
-
I have 60 yrs of hourly 3km resolution dynamically downscaled ESM projections and climate reanalysis for NE US along with coarse driver data if your team is interested ( ~2 petabytes)
-
Reached out via email. Thanks!
Kraj razgovora
Novi razgovor -
-
-
Thank you! however I’m still waiting to hear about a breakthrough in translation from dolphin to human languages using AI ;-)
-
We're still trying to give our models a sense of porpoise before we tackle the derivative problem you mention.
- Još 1 odgovor
Novi razgovor -
-
-
@JeffDean could you add detail on climate data sources/backend used at@GoogleAI ? A free climate & economics data infrastructure would be a great way to support projects like@ClimateChangeAI in particular and#ClimateAction in general. Looking forward to it!#neurips2019 -
For this precipitation forecasting work, we're using the MRMS dataset from NOAA (see section 2 of https://arxiv.org/abs/1912.12132 for details). There are lots of datasets in
@googleearth Engine. You can browse them at https://developers.google.com/earth-engine/datasets/tags … (in particular, see ones tagged as 'climate') - Još 1 odgovor
Novi razgovor -
-
-
A question off this: Have you all been exploring using machine learning to emulate general circulation models, which are very expensive to simulate? I feel like the ML machinery for solving ODEs, PDEs, and respecting symmetries etc., is now mature enough to attack those problems.
-
Gaussian process emulators are a great method for emulating GCMs - I'm currently doing a PhD involving emulating HadCM3 with promising results.
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.