LinUCB
A classic algorithm for contextual bandits that assumes a linear relationship between expected reward and contextual features. It maintains parameter estimates for each arm to compute an upper confidence bound, and preferentially selects the arm with the highest bound to balance exploration and exploitation.
Timeline (last 90 days)
Common algorithmic implementations of contextual bandits include LinUCB (Linear Upper Confidence Bound) and Thompson Sampling.
LinUCB assumes that the expected reward has a linear relationship with contextual features, computes the upper confidence bound by maintaining parameter estimates for each arm, and prioritizes the option with the highest upper bound.
LinUCB and Thompson Sampling generally perform more stably in practice than a fixed ε-greedy strategy.
All Facts (3)
Common algorithmic implementations of contextual bandits include LinUCB (Linear Upper Confidence Bound) and Thompson Sampling.
50%UnverifiedLinUCB assumes that the expected reward has a linear relationship with contextual features, computes the upper confidence bound by maintaining parameter estimates for each arm, and prioritizes the option with the highest upper bound.
50%UnverifiedLinUCB and Thompson Sampling generally perform more stably in practice than a fixed ε-greedy strategy.
50%