3 ms·
Well this is getting in some rather unfamiliar territory to me, but a function like max(0,x) could be considered a polynomial in the max-plus algebra [1]. In th
by contravariant 8y ago
Well this is getting in some rather unfamiliar territory to me, but a function like max(0,x) could be considered a polynomial in the max-plus algebra [1]. In this algebra the values kind of behave like the base-b log of a real number, with b taken to infinity [2]. If you replace b by a really big number you're just dealing with regular polynomials.
In this sense I think you could consider a big neural network with RELU activation a (really big) polynomial.
This is about the extent of my knowledge on the subject though, for more info search for tropical geometry [3].
[1]: https://en.wikipedia.org/wiki/Max-plus_algebra https://en.wikipedia.org/wiki/Max-plus_algebra
[2]: https://mathoverflow.net/questions/83624/why-tropical-geometry https://mathoverflow.net/questions/83624/why-tropical-geomet...
[3]: https://en.wikipedia.org/wiki/Tropical_geometry https://en.wikipedia.org/wiki/Tropical_geometry
- sdenton4 8y agoTotally. I'm unsure, though, how much people have looked at regression in tropical geometry, though... There's likely not a whole lot of results to draw on. (Similarly, I looked real hard at hyperplane arrangements arising from relu activations at one point, and found the hyper plane arrangement literature to be pay much completely useless for statistical questions... Though it's still quite beautiful!)
- rors 8y agoI wrote my thesis on tropical geometry for machine translation[1]. You can use it to train a a linear model for structured prediction, which used to be the dominant model before RNNs. [1]: http://www.aclweb.org/anthology/N15-1041 http://www.aclweb.org/anthology/N15-1041