Blind Network Revenue Management and Bandits with Knapsacks under Limited Switches
This work addresses resource-constrained online learning for revenue management, with incremental contributions to handling limited switches in bandit problems.
The paper tackles the problem of dynamic pricing with demand learning under resource and switching constraints, establishing matching upper and lower bounds on optimal regret and developing efficient algorithms that achieve it, with simulations showing strong reward performance and reduced switches.
This paper studies the impact of limited switches on resource-constrained dynamic pricing with demand learning. We focus on the classical price-based blind network revenue management problem and extend our results to the bandits with knapsacks problem. In both settings, a decision maker faces stochastic and distributionally unknown demand, and must allocate finite initial inventory across multiple resources over time. In addition to standard resource constraints, we impose a switching constraint that limits the number of action changes over the time horizon. We establish matching upper and lower bounds on the optimal regret and develop computationally efficient limited-switch algorithms that achieve it. We show that the optimal regret rate is fully characterized by a piecewise-constant function of the switching budget, which further depends on the number of resource constraints. Our results highlight the fundamental role of resource constraints in shaping the statistical complexity of online learning under limited switches. Extensive simulations demonstrate that our algorithms maintain strong cumulative reward performance while significantly reducing the number of switches.