1 paper · 1 filter
Einar Urdshals, Edmund Lau, Jesse Hoogland +2
We study neural network compressibility by using singular learning theory to extend the minimum description length (MDL) principle to singular models like neural networks. Through…