Interesting! I was under the impression that ReLU was recent breakthrough? Nair & Hinton 2010 https://www.cs.toronto.edu/~fritz/absps/reluICML.pdf …
-
-
-
Hinton was a co-author of the chapter this is from :)
- Još 9 drugih odgovora
Novi razgovor -
-
-
Those books! I still love reading them.pic.twitter.com/CM5tEQKOGM
-
I've never read vol 2. Anything of interest to a modern reader?
- Još 2 druga odgovora
Novi razgovor -
-
-
I remember buying that book and trying to train networks using the SW that came with ut in 1989 No fun when you had to create your own data and do the training on a crappy olivetti 8086! Thankfully things have moved on since then!
-
Yeah I didn't get much out of the book back then either. Dr Dobbs had stuff I could use in 1989 though!
Kraj razgovora
Novi razgovor -
-
-
My understanding is that threshold functions are general, you can define (or find) the threshold and slope. RELUs are a special type of threshold functions, where the threshold is zero, and the slope is one. I might be wrong though, this is coming from the sysID field.
-
There are a lot of different ReLUs, including PReLU (parameterized) and LReLU (leaky).
Kraj razgovora
Novi razgovor -
-
-
In one sense, "rectified linear unit" is more descriptive about the underlying function, while "threshold function" is less so. For instance, by looking at the term alone, you could mistake it for a step function.
-
Yes you could.
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.