Timeline for Modeling Pseudorandom Number Generator Without Replacement
Current License: CC BY-SA 4.0
3 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 19, 2020 at 9:31 | comment | added | Henry | @dhakim No - if the probability that any of your individual bits is zero or one is 50-50 then the expected number of correct guesses out of $100$ can never be better than $50$. But what can change with different strategies is the distribution of correct guesses: the second strategy increases the probability of exactly $60$ or $40$ correct guesses but reduces the probability of $59$ or $41$ correct guesses while the third does the opposite | |
| Jul 19, 2020 at 2:09 | comment | added | dhakim | Does this mean that a clever adversary (or in the worst case, my adversarial black box machine learning classifier), can do slightly better than one would predict with either the binomial or hypergeometric distributions if they can figure out that my target set is always a shuffled deck of 0's and 1's? | |
| Jul 18, 2020 at 22:34 | history | answered | Henry | CC BY-SA 4.0 |