|
|
|
|
|
|
by binarycoffee
2170 days ago
|
|
|
So this is not what the article is about. It discusses a new method to generate integer values according to an arbitrary probability distribution, using as input a uniform random generator. Whether the input generator is truly random, pseudorandom, cryptographically secure etc, is irrelevant: the output will presumably only be as random/secure as the input RNG. Admittedly, the article does a poor job at explaining what the FLDR method is about, and it looks as biased as the method itself (sorry for the easy pun).
From my understanding of the paper [1], the method is better than the state-of-the-art Alias method [2] only when the entropy of the target distribution is low. When entropy is high, it performs similarly (or may even be a bit slower) but uses up to 8 times more memory space. [1] https://arxiv.org/abs/2003.03830 [2] https://en.wikipedia.org/wiki/Alias_method |
|
|