Errrm @SimonDeDeo did say deep learning was an "amazing feat of engineering." @RandomlyWalking took multiple tweets to basically concur. Not sure who is misconstruing what.
-
-
-
Misconstruing engineering as being not research
-
He didn’t actually say that, he said ML is a feat in engineering but it’s not science. Big difference.
-
I'd be happy indeed if I were misinterpreting him, but he literally said that there are no new ideas in ML since the 1990s and that it is impossible to do cool research at Google.
-
Hard to reconcile those two statements with a more nuanced statement like, "The best research in engineering is technically changes and introduces new concepts that provide insight upstream into maths and science"
-
Engineering for a corporation is not science.
-
Sometimes an engineering project in a corporation needs new science to achieve its goals. https://en.wikipedia.org/wiki/History_of_the_transistor …
-
They should probably kick more of that back to the state University system.
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
This response contain a lot of debunked myths on how generalization work in ML. Please stop spreading them.
-
I see only one mention of regularization: that the theory says DL should not work due to the curse of dimensionality, but it does much better than the theory predicts w/o any (explicit) regularisation. Can you be more specific what he says is wrong and debunked, please?
-
Or you wanted to say that we always knew that constraining the norm of the model is just one way to regularise models, and that we knew long before DL that the model structure acts as regulariser?
-
We know much more than that! E.g. Number of parameters can be infinite, SGD gives small L2 norms, complexity of class not important. See this old thread here:https://twitter.com/alirahimi0/status/956242762772262912?s=19 …
-
Thanks for clarifying!
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
That's because Charles is a nice dude.
- কথা-বার্তা শেষ
নতুন কথা-বার্তা -
লোড হতে বেশ কিছুক্ষণ সময় নিচ্ছে।
টুইটার তার ক্ষমতার বাইরে চলে গেছে বা কোনো সাময়িক সমস্যার সম্মুখীন হয়েছে আবার চেষ্টা করুন বা আরও তথ্যের জন্য টুইটারের স্থিতি দেখুন।
(post 1 of "until my beer gets cold")


