pith. sign in

arxiv: 1706.03301 · v1 · pith:HJN3TP7Fnew · submitted 2017-06-11 · 💻 cs.LG · cs.NE· stat.ML

Neural networks and rational functions

classification 💻 cs.LG cs.NEstat.ML
keywords rationalepsilonrelufunctionfunctionsnetworktextapproximate
0
0 comments X
read the original abstract

Neural networks and rational functions efficiently approximate each other. In more detail, it is shown here that for any ReLU network, there exists a rational function of degree $O(\text{polylog}(1/\epsilon))$ which is $\epsilon$-close, and similarly for any rational function there exists a ReLU network of size $O(\text{polylog}(1/\epsilon))$ which is $\epsilon$-close. By contrast, polynomials need degree $\Omega(\text{poly}(1/\epsilon))$ to approximate even a single ReLU. When converting a ReLU network to a rational function as above, the hidden constants depend exponentially on the number of layers, which is shown to be tight; in other words, a compositional representation can be beneficial even for rational functions.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.