Follow

AI rant 

If I were to make a book about the most influential theorems of the 1900s, this would definitely make it in the book.

"Approximation by superpositions of a sigmoidal function"

The possible use of signal processing and control applications, is basically the same as deep learning. Later papers show that deep learning is even less limited. But think, all of modern computers and electronics, live in a subset of what machine learning can simulate.

api.semanticscholar.org/Corpus

AI rant 

@jmw150 yeah, it's important, but I'm not sure how useful it is for machine learning since you can't necessarily train a learner to produce that function approximation in a reasonable timespan

AI rant 

@2ck

It was the start of a dream really.

There are many other papers published that expanded on this one, even now in 2021. It is the start of machine learning actually being considered practical. Later papers compare convergence rates and error bounds between models, along with limitations on representatbility.

Before this it was an AI winter. No one took neurons seriously after it was demonstrated XOR was not even representable, with a single layer. Almost everything had to be directly built, as so progress was limited.

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.