Comment by datastoat

I like Bayesian inference for few-parameter models where I have solid grounds for choosing my priors. For neural networks, I like to ask people "what's your prior for ReLU versus LeakyReLU versus sigmoid?" and I've never gotten a convincing answer.