« on: December 12, 2016, 03:25:48 PM »
The rub on machine learning for 16/17 freezing is that unless someone made it a big project I don't see it being that useful, perhaps only as good as forecasting ice disposition a few days into the future. With how unprecedented the past few weeks have been, I'd also be wary of a model that relied too much on historic data since it would have a hard time accounting for what we are seeing now.
Its a shame the NIPS slides didn't cover any deep learning, its also a shame that NIPS'16 just ended and I can't find anything about climate modeling in the presentations. Archimid, don't give AlphaGo too much credit, the paper google released on it is a bit technical but the entire system is pretty straightforward to understand. As with all things deep learning, the end result will depend on the quality and quantity of data available.