1 paper
Tjeerd Jan Heeringa, Len Spek, Felix Schwenninger +1
The approximation properties of infinitely wide shallow neural networks heavily depend on the choice of the activation function. To understand this influence, we study embeddings b…