Linear Bandits with Limited Adaptivity and Learning Distributional Optimal Design
About
Motivated by practical needs such as large-scale learning, we study the impact of adaptivity constraints to linear contextual bandits, a central problem in online active learning. We consider two popular limited adaptivity models in literature: batch learning and rare policy switches. We show that, when the context vectors are adversarially chosen in $d$-dimensional linear contextual bandits, the learner needs $O(d \log d \log T)$ policy switches to achieve the minimax-optimal regret, and this is optimal up to $\mathrm{poly}(\log d, \log \log T)$ factors; for stochastic context vectors, even in the more restricted batch learning model, only $O(\log \log T)$ batches are needed to achieve the optimal regret. Together with the known results in literature, our results present a complete picture about the adaptivity constraints in linear contextual bandits. Along the way, we propose the distributional optimal design, a natural extension of the optimal experiment design, and provide a both statistically and computationally efficient learning algorithm for the problem, which may be of independent interest.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Linear Contextual Bandits | Synthetic Linear Bandit K=5000, d=10 | Average Runtime (s)555.1 | 12 | |
| Linear Contextual Bandits | Synthetic Linear Bandit K=1000, d=5 | Average Runtime (s)290.9 | 6 | |
| Linear Contextual Bandits | Synthetic Linear Bandit K=50, d=20 | Average Runtime (s)1.03e+3 | 6 | |
| Linear Contextual Bandits | Synthetic Linear Bandit (K=100, d=30) | Average Runtime (s)2.99e+3 | 6 | |
| Linear Contextual Bandits | Normal Contexts K=1000, d=5 synthetic (test) | Average Runtime (s)148.2 | 6 | |
| Linear Contextual Bandits | Normal Contexts K=50, d=20 synthetic (test) | Average Runtime (s)981.2 | 6 | |
| Linear Contextual Bandits | Normal Contexts K=100, d=30 synthetic (test) | Average Runtime (s)2.77e+3 | 6 |