7 ms·
That’s a pretty useless “proof” which gives no insight into how neural networks learn in practice. The author takes a sigmoid neuron, tweaks its parameters to t
by aaaaaaaaaaab 5y ago
That’s a pretty useless “proof” which gives no insight into how neural networks learn in practice. The author takes a sigmoid neuron, tweaks its parameters to the extreme so that it looks like a step function, then concludes that since a linear combination of step functions can approximate anything, so do neural networks. Bravo.
- r-zip 5y agoDid you ever consider that explaining how deep nets "learn in practice" is not the point?
- anothernewdude 5y agoIf it's not about learning this result is useless. A lookup table can compute any function, but who cares?
- mdp2021 5y ago> who cares The people who need to build some (sophisticated kind of) lookup table for a function of unavailable details through automation.