The Impact of Batch Learning in Stochastic Linear Bandits
This work addresses batch-centric scenarios in bandit problems, which is incremental but practically relevant for sequential decision-making applications.
The paper tackles the problem of batch learning in stochastic linear bandits, showing that batch learning introduces a multiplicative factor of batch size relative to online regret, with theoretical bounds validated empirically.
We consider a special case of bandit problems, named batched bandits, in which an agent observes batches of responses over a certain time period. Unlike previous work, we consider a more practically relevant batch-centric scenario of batch learning. That is to say, we provide a policy-agnostic regret analysis and demonstrate upper and lower bounds for the regret of a candidate policy. Our main theoretical results show that the impact of batch learning is a multiplicative factor of batch size relative to the regret of online behavior. Primarily, we study two settings of the stochastic linear bandits: bandits with finitely and infinitely many arms. While the regret bounds are the same for both settings, the former setting results hold under milder assumptions. Also, we provide a more robust result for the 2-armed bandit problem as an important insight. Finally, we demonstrate the consistency of theoretical results by conducting empirical experiments and reflect on optimal batch size choice.