Learning Machine Learning

[$alt]

Machine learning (ML) :
The study of computer algorithms that can improve automatically through experience and by the use of data.[1] It is seen as a part of artificial intelligence. Machine learning algorithms build a model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so.[2] Machine learning algorithms are used in a wide variety of applications, such as in medicine, email filtering, speech recognition, and computer vision, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.

What do you guys think this means for weather forecasting?

 

[GaWx]

Machine learning (ML) :
The study of computer algorithms that can improve automatically through experience and by the use of data.[1] It is seen as a part of artificial intelligence. Machine learning algorithms build a model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so.[2] Machine learning algorithms are used in a wide variety of applications, such as in medicine, email filtering, speech recognition, and computer vision, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.

What do you guys think this means for weather forecasting?

Interesting topic. So, would ML allow wx models to adjust for their initial biases and later reduce, if not eliminate, them without an actual model upgrade? Has ML yet been tried for wx models? Also, would the met. profession want/allow for this to happen knowing that better models may later mean less need for forecasters?

 

[$alt]

Here's an article exploring titled:
"Can Deep Learning Beat Numerical Weather Prediction?"

Can deep learning beat numerical weather prediction? | Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences

The recent hype about artificial intelligence has sparked renewed interest in applying the successful deep learning (DL) methods for image recognition, speech recognition, robotics, strategic games and other application areas to the field of meteorology. ...

royalsocietypublishing.org

 

[pcbjr]

Shawn,
Funny you raise this, as this afternoon I was pondering how models have missed so much of what was "supposed" to happen this year in a Nina. I hypothecated that something has changed from historical "norms" (albeit we have but a brief history of those) and perhaps to some extent the models are relying on erroneous input leading to erroneous output. I wondered if it is possible that a model (if constructed correctly) might "learn" from its past mistakes based on more current data and "patterns" as opposed to weighing and producing output in favor of data that may not correlate with current cause and effect? The old GIGO thing.
Then you post this.
Now I'm really curious.

Phil

 

[weatherfide]

Here's an article exploring titled:
"Can Deep Learning Beat Numerical Weather Prediction?"

Can deep learning beat numerical weather prediction? | Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences

The recent hype about artificial intelligence has sparked renewed interest in applying the successful deep learning (DL) methods for image recognition, speech recognition, robotics, strategic games and other application areas to the field of meteorology. ...

royalsocietypublishing.org

I skimmed through that. It's interesting and I kind of see where they are heading, but I still don't know if you can overcome the missing observational data. There is already considerable error in the inputs, regardless of what method is being used.

 

[atlwx]

Shawn,
Funny you raise this, as this afternoon I was pondering how models have missed so much of what was "supposed" to happen this year in a Nina. I hypothecated that something has changed from historical "norms" (albeit we have but a brief history of those) and perhaps to some extent the models are relying on erroneous input leading to erroneous output. I wondered if it is possible that a model (if constructed correctly) might "learn" from its past mistakes based on more current data and "patterns" as opposed to weighing and producing output in favor of data that may not correlate with current cause and effect? The old GIGO thing.
Then you post this.
Now I'm really curious.

Phil

It sounds so easy. But even if you train some ML model to “learn the mistakes” of a weather model (something not trivial to do)… you cant just look at the model (machine learning generated one, not the wx model) and easily deduce what “the mistakes” actually are.

 

[pcbjr]

It sounds so easy. But even if you train some ML model to “learn the mistakes” of a weather model (something not trivial to do)… you cant just look at the model (machine learning generated one, not the wx model) and easily deduce what “the mistakes” actually are.

Never inferred you could ... just courious about the phantom Hal the Weather Computer ...

 

[atlwx]

Never inferred you could ... just courios about the phantom Hal the Weather Computer ...

I’ve always dreamed of the weather predicting possibilities since the first time I dinked around with back props (type of neural network) in Matlab years and years (so many years) ago.