Sample Efficient Policy Search for Optimal Stopping Domains
This addresses sample efficiency in optimal stopping problems, which is incremental but improves theoretical guarantees and practical performance.
The paper tackles the problem of simultaneously learning and planning in optimal stopping domains with limited data, proposing GFSE, a model-free policy search method that reuses data for sample efficiency. The result includes a tightened PAC bound with logarithmic dependence on horizon length and empirical benefits over existing methods on three diverse domains.
Optimal stopping problems consider the question of deciding when to stop an observation-generating process in order to maximize a return. We examine the problem of simultaneously learning and planning in such domains, when data is collected directly from the environment. We propose GFSE, a simple and flexible model-free policy search method that reuses data for sample efficiency by leveraging problem structure. We bound the sample complexity of our approach to guarantee uniform convergence of policy value estimates, tightening existing PAC bounds to achieve logarithmic dependence on horizon length for our setting. We also examine the benefit of our method against prevalent model-based and model-free approaches on 3 domains taken from diverse fields.