While reading on randomization for online learning from the text "Online Learning and Online Convex Optimization" by Shai Shalev-Shwartz, I found the following statement (Pg 115)- "The adversary can still know the learner’s forecasting strategy and even the random bits of previous rounds, but it doesn’t know the actual value of the random bits used by the learner on round t."
Assume the learner is using a coin with bias "p" for the random experiment. Does the statement mean that it precludes the adversary from knowing the bias "p" of the coin as well?
My guess is it does not, I assumed that the adversary just does not know the outcome of the coin flip but can know the bias. Can someone clarify.
