Chapter 7, Decoupling Exploration and Exploitation in Multi-Armed Bandits

1. The true probability distribution of an event is the actual likelihood of encountering that event. We discover this distribution through repeated experimental trials.
2. Conducting more trials gives us a better picture of the true probability distribution of a problem. In most problem spaces, conducting 10 trials of an event would not give us sufficient data to develop a detailed model of the event.
3. A small sample size might be biased in a way the experimenter is not aware of, and the descriptive statistics of that sample might not be reflected in a larger sample.
4. Thompson sampling is a Bayesian method for optimization that involves choosing a prior probability distribution for an event and updating that as more information about the event is received. Since it is computationally expensive to try to find the...