We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Quasi-optimal Learning with Continuous Treatments

Abstract: Many real-world applications of reinforcement learning (RL) require making decisions in continuous action environments. In particular, determining the optimal dose level plays a vital role in developing medical treatment regimes. One challenge in adapting existing RL algorithms to medical applications, however, is that the popular infinite support stochastic policies, e.g., Gaussian policy, may assign riskily high dosages and harm patients seriously. Hence, it is important to induce a policy class whose support only contains near-optimal actions, and shrink the action-searching area for effectiveness and reliability. To achieve this, we develop a novel \emph{quasi-optimal learning algorithm}, which can be easily optimized in off-policy settings with guaranteed convergence under general function approximations. Theoretically, we analyze the consistency, sample complexity, adaptability, and convergence of the proposed algorithm. We evaluate our algorithm with comprehensive simulated experiments and a dose suggestion real application to Ohio Type 1 diabetes dataset.
Comments: The first two authors contributed equally to this work
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Methodology (stat.ME)
Cite as: arXiv:2301.08940 [stat.ML]
  (or arXiv:2301.08940v1 [stat.ML] for this version)

Submission history

From: Wenzhuo Zhou [view email]
[v1] Sat, 21 Jan 2023 11:30:13 GMT (814kb,D)

Link back to: arXiv, form interface, contact.