I would like to address the following question. How many samples is required to distinguish between a Bernoulli distribution with parameter $1/2 - \epsilon$ and a uniform distribution over $\{0, 1\}$. I know that this statement is not mathematically rigorous, so I'm looking to answer two questions:
- How can I rephrase the statement to be mathematically rigorous?
- I feel like somehow concentration measures (like Hoeffding's inequality) may help to answer the question, but don't know how, any idea?