Re:Why deep learning works even though it shouldn’t
by dnautics on Tuesday October 20, 2020 @07:34PM
(#24843381)
> if you have enough lines you can approximate any curve within a margin. This is what large neural networks are doing.
No, this is actually exactly the opposite of what the article is saying.
"If you have enough lines you can approximate any curve within a margin" is usually a bad thing because your approximations lose meaning as the number of lines increase.
The surprising thing about machine learning is that increasing the "number of lines", if you will, increases the meaning, too, and there's some wierd and subtle properties about mathematics in higher dimensions that makes this work.
No, this is actually exactly the opposite of what the article is saying.
"If you have enough lines you can approximate any curve within a margin" is usually a bad thing because your approximations lose meaning as the number of lines increase.
The surprising thing about machine learning is that increasing the "number of lines", if you will, increases the meaning, too, and there's some wierd and subtle properties about mathematics in higher dimensions that makes this work.