Contextual Bandits with Linear Payoff Functions

This paper introduces and analyzes a contextual bandit algorithm called LinUCB (Linear Upper Confidence Bound). LinUCB uses linear regression to estimate the expected reward for each action (or in Dopt's case, onboarding flow variation) given the user's context. It balances exploration (trying new flows) and exploitation (showing the flow predicted to perform best) by using an upper confidence bound on the reward estimate. This exploration-exploitation trade-off is crucial for optimizing user onboarding flows over time based on user behavior, a key functionality of Dopt.