Presentation: Reinforcement Learning Strategies in Experimentation

Talk:

A/B testing or Experimentation plays a key role in decision making. Data driven approaches with science underpinnings provide the needed rigor to test hypotheses and make decisions. There are however a class of decisions where variable assignments based on Reinforcement Learning techniques make some tests feasible and add velocity (shorter experimentation duration) to others. These strategies don’t conform to the standard hypothesis testing template but serve the same end goal of data driven decision making. We will go through an overview of multi-arm bandits and bayesian parameter tuning as RL strategies applied to experimentation systems.