Multi-Objective SPIBB: Seldonian Offline Policy Improvement with Safety Constraints in Finite MDPs

Abstract

We study the problem of Safe Policy Improvement (SPI) under constraints in the offline Reinforcement Learning (RL) setting. We consider the scenario where: (i) we have a dataset collected under a known baseline policy, (ii) multiple reward signals are received from the environment inducing as many objectives to optimize. We present an SPI formulation for this RL setting that takes into account the preferences of the algorithm's user for handling the trade-offs for different reward signals while ensuring that the new policy performs at least as well as the baseline policy along each individual objective. We build on traditional SPI algorithms and propose a novel method based on Safe Policy Iteration with Baseline Bootstrapping (SPIBB, Laroche et al., 2019) that provides high probability guarantees on the performance of the agent in the true environment. We show the effectiveness of our method on a synthetic grid-world safety task as well as in a real-world critical care context to learn a policy for the administration of IV fluids and vasopressors to treat sepsis.

References in corpus (15)

• Deep reinforcement learning from human preferences
• CVXPY: A Python-Embedded Modeling Language for Convex Optimization
• Challenges of Real-World Reinforcement Learning
• Empirical Bernstein Bounds and Sample Variance Penalization
• Scalable agent alignment via reward modeling: a research direction
• AI Safety Gridworlds
• A Generalized Algorithm for Multi-Objective Reinforcement Learning and Policy Adaptation
• Safe Policy Improvement with Baseline Bootstrapping
• Doubly Robust Off-policy Value Evaluation for Reinforcement Learning
• A Lyapunov-based Approach to Safe Reinforcement Learning
• Safe Policy Improvement by Minimizing Robust Baseline Regret
• Environment-Independent Task Specifications via GLTL
• A Distributional View on Multi-Objective Policy Optimization
• Clinician-in-the-Loop Decision Making: Reinforcement Learning with Near-Optimal Set-Valued Policies
• Trajectory Inspection: A Method for Iterative Clinician-Driven Design of Reinforcement Learning Studies