#66 Machine Learning & Data Science Challenge 66

How to initialize biases in deep learning? It is possible and common to initialize the biases to be zero since the random numbers in the weights provide the asymmetry braking. For ReLU non-linearities, some people like to use small constant values ...