Microsoft Says Its Aurora AI Can Accurately Predict Air Quality, Typhoons (techcrunch.com) 28
An anonymous reader quotes a report from TechCrunch: One of Microsoft's latest AI models can accurately predict air quality, hurricanes, typhoons, and other weather-related phenomena, the company claims. In a paper published in the journal Nature and an accompanying blog post this week, Microsoft detailed Aurora, which the tech giant says can forecast atmospheric events with greater precision and speed than traditional meteorological approaches. Aurora, which has been trained on more than a million hours of data from satellites, radar and weather stations, simulations, and forecasts, can be fine-tuned with additional data to make predictions for particular weather events.
AI weather models are nothing new. Google DeepMind has released a handful over the past several years, including WeatherNext, which the lab claims beats some of the world's best forecasting systems. Microsoft is positioning Aurora as one of the field's top performers -- and a potential boon for labs studying weather science. In experiments, Aurora predicted Typhoon Doksuri's landfall in the Philippines four days in advance of the actual event, beating some expert predictions, Microsoft says. The model also bested the National Hurricane Center in forecasting five-day tropical cyclone tracks for the 2022-2023 season, and successfully predicted the 2022 Iraq sandstorm.
While Aurora required substantial computing infrastructure to train, Microsoft says the model is highly efficient to run. It generates forecasts in seconds compared to the hours traditional systems take using supercomputer hardware. Microsoft, which has made the source code and model weights publicly available, says that it's incorporating Aurora's AI modeling into its MSN Weather app via a specialized version of the model that produces hourly forecasts, including for clouds.
AI weather models are nothing new. Google DeepMind has released a handful over the past several years, including WeatherNext, which the lab claims beats some of the world's best forecasting systems. Microsoft is positioning Aurora as one of the field's top performers -- and a potential boon for labs studying weather science. In experiments, Aurora predicted Typhoon Doksuri's landfall in the Philippines four days in advance of the actual event, beating some expert predictions, Microsoft says. The model also bested the National Hurricane Center in forecasting five-day tropical cyclone tracks for the 2022-2023 season, and successfully predicted the 2022 Iraq sandstorm.
While Aurora required substantial computing infrastructure to train, Microsoft says the model is highly efficient to run. It generates forecasts in seconds compared to the hours traditional systems take using supercomputer hardware. Microsoft, which has made the source code and model weights publicly available, says that it's incorporating Aurora's AI modeling into its MSN Weather app via a specialized version of the model that produces hourly forecasts, including for clouds.
Great! (Score:4, Insightful)
A new software Nostradamus is born, from the company that's known for the safest, most reliable and bugfree software ever.
Re: (Score:1)
Rules of thumb... (Score:1)
...the tech giant says can forecast atmospheric events with greater precision and speed than traditional meteorological approaches...
Yeah, but we don't know exactly the mechanism these models use. For all we know, it's a rule-of-thumb that often works, but breaks when something critically important needs to be actually accurate.
Re: Rules of thumb... (Score:1)
Isn't it an attention mechanism, counting co-occurrences of phenomena at long range in time and space?
Re: (Score:2)
AI doesn't use rules, normal models use those, and they already do break when something critically important needs to be accurate.
Re: (Score:2)
For all we know, it's a rule-of-thumb that often works,
Neural networks are basically all approximations of a curve in multidimensional space. If your dataset were small enough (or your computer were powerful enough), then you would just use Bayesian classification and get an exact answer.
Re: Rules of thumb... (Score:1)
Why were neural networks alone unable to provide the context sensitivity that the attention mechanism delivers?
Re: (Score:2)
As to the answer to your specific question, I don't know the answer.
Re: (Score:2)
Because they can not "just be trained".
You need a network architecture that suits the problem.
And for many problems: that architecture is unknown.
Re: (Score:2)
Better a rule of thumb than nothing.
Hmm (Score:1)
Re: (Score:1)
Bill Gates didn't suck off Trump, though.
Re:Hmm (Score:4, Informative)
Yes, that was a job he left to his CEO.
Re: (Score:2)
Oh, good point.
Yeah, but they had a common friend (Score:2)
Though one of them had the good sense to stop haning around with that guy once it became clear he was bad news.
Re:Hmm (Score:4, Informative)
Not really. Unless the tech company wants to invest in weather satellites showing visible and IR weather patterns, aircraft to fly through hurricanes to gather pressure and wind speed data, ships to deploy ocean buoys to gather water temperature, and Doppler radar sites to gather ground weather data.
The National Weather Service (NWS) does more than just make predictions. They also gather all the data to feed into the weather models and archive the data for later analysis to improve their prediction models. Add in tsunami warning, waterway mapping, fishery monitoring, marine mammal protection and general marine science and NOAA does a pretty vital job with only 12,000 employees.
If Fred&Barney's AI shop says that using data from the NWS, their predictions are more accurate 50% of the time, then they are just guessing and playing the averages.
Re: (Score:3)
Asked Gemini and it's pretty much everyone needs government as a basis for anything weather. [g.co]
Re: (Score:2)
The history engine has a problem (Score:2)
As climate change affects the fluid dynamics of the atmosphere and the oceans, all that history will become worthless
Re: (Score:2)
Book-banning is the Quigley/Hayes Code: No nudity, sex, homosexuality, racism, or drugs, only immigrant criminals.
Have you watched youtube lately? It censors a lot more than the Hayes code ever did.
WeatherNews is the most accurate (Score:2)
It's a fact that every one knows to be true like WinAmp whipping the llama's ass.
Accuracy And Range Of Vision (Score:2)
Accurate landfall prediction 4 days in advance is unimpressive by today's standards. In the last few years, we have repeatedly seen hurricane predictions with remarkable accuracy of path, landfall location, strength fluctuations, as early as 10 days in advance. Four days a a sad relic of prior decades.
Call it AI, or models, or whatever you want, but the currently used predictions are amazing. I think that the most accurate ones that I've seen are coming from Barron Weather. But, I'm not positive about this.
air quality? (Score:2)