Most state of art algorithms right now is using/exploiting big data. My concern is what can you do to maximize reward while waiting for large amount of data that you think is appropriate.
For example, in predicting if a customer will churn (the reward is the revenue from all customers), the usual practice is to wait for data to accumulate. Is there a way to exploit the incoming information before exploiting the big data in full blown machine learning?
An easier example: you want to identify A over B and vice versa. The reward will be the accurate prediction which will be given by the user.
I know it is reinforcement learning problem but how will you implement it? I know that an on line training of ML algorithm will not work because you don't even know the algorithm that is suitable for the problem.