Kaarel comments on A circuit prior in NN-bayes

[Kaarel]

the short answer is: yes for bayesian learning, no for gradient descent