3 ms·
it can't learn exponents. it can only multiply. there's also a difference between memorizing part of a line, and being able to extrapolate it.
by Kalanos 5y ago
it can't learn exponents. it can only multiply.
there's also a difference between memorizing part of a line, and being able to extrapolate it.
- iamcurious 5y agoI find this very interesting, can you expand or provide links?
- Kalanos 5y agoweights that connect neurons can only be multiplied. now i am looking at activation functions like `exponent`. i just doubt a neural network regression would be able to predict rlly high values accurately for something like E = mc^2.
- iamcurious 5y agothank you.
- hlbjhblbljib 5y agoThey can with the right activation function. I'm thinking inverse log.