Degrees of Freedom in Deep Neural Networks

Tianxiang Gao, Vladimir Jojic

In this paper, we explore degrees of freedom in deep sigmoidal neural networks. We show that the degrees of freedom in these models is related to the expected optimism, which is th…