5 ms·
When it comes to machine learning, yes it is important to know statistical methods, but at this point deep learning shouldn't be pejoratively referred to as the
by typon 5y ago
When it comes to machine learning, yes it is important to know statistical methods, but at this point deep learning shouldn't be pejoratively referred to as the HottestNewArchitecture. If you want to solve previously unsolved problems, deep learning is necessary; and if you want more specific advice, learn how transformers work. This class of models is qualitatively different in the kinds of problems they unlock as solvable.
- softwaredoug 5y agoI don't disagree. But I think (a) Important stepwise changes in a domain are actually rarer than we think. Transformers & deep learning being one good example. (b) Focusing on the background information prepares you for that stepwise change, and recognize the stepwise change when it happens Instead of trying to suss out from the tech literatti every new thing and chase that, I'd rather spend my time on "old things" and be better prepared for the right stepwise changes. There's a lot of 'revolutionary' and 'big ideas' that in reality are marketing and/or repackaging old ideas.