Randomized smoothing is one of the most promising techniques for adversarial training with guarantees. This work further generalizes randomized techniques with new methods, overcoming previous impossibility results for randomized smoothing with certain norms. This work is likely to be very influential among people studying adversarial machine learning.