Corpus ID: 222272185

Trajectory Inspection: A Method for Iterative Clinician-Driven Design of Reinforcement Learning Studies

@article{Ji2020TrajectoryIA,
  title={Trajectory Inspection: A Method for Iterative Clinician-Driven Design of Reinforcement Learning Studies},
  author={Christina X. Ji and Michael Oberst and Sanjat Kanjilal and D. Sontag},
  journal={ArXiv},
  year={2020},
  volume={abs/2010.04279}
}
  • Christina X. Ji, Michael Oberst, +1 author D. Sontag
  • Published 2020
  • Computer Science
  • ArXiv
  • Treatment policies learned via reinforcement learning (RL) from observational health data are sensitive to subtle choices in study design. We highlight a simple approach, trajectory inspection, to bring clinicians into an iterative design process for model-based RL studies. We inspect trajectories where the model recommends unexpectedly aggressive treatments or believes its recommendations would lead to much more positive outcomes. Then, we examine clinical trajectories simulated with the… CONTINUE READING

    Figures from this paper.

    References

    SHOWING 1-10 OF 24 REFERENCES
    Counterfactual Off-Policy Evaluation with Gumbel-Max Structural Causal Models
    • 34
    • PDF
    Continuous State-Space Models for Optimal Sepsis Treatment: a Deep Reinforcement Learning Approach
    • 78
    • PDF
    Identifying Distinct, Effective Treatments for Acute Hypotension with SODA-RL: Safely Optimized Diverse Accurate Reinforcement Learning
    • 2
    • PDF
    The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care
    • 225
    Model-Based Reinforcement Learning for Sepsis Treatment
    • 10
    • PDF
    Off-Policy Deep Reinforcement Learning without Exploration
    • 151
    • PDF