Pith. sign in

REVIEW 2 major objections 2 minor 45 references

Interaction-Limited Safe Continuous-Time RL for Dynamical Medical Treatment

T0 review · 2 major / 2 minor · reviewed 2026-06-28 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Suitably constructed constraints at clinical interaction times guarantee safety over the full continuous-time patient trajectory with high probability.

desk verdict The paper gives a concrete way to optimize both treatment and interaction timing in continuous-time medical RL with safety that transfers from discrete points to full trajectories, but the guarantees rest on the SMDP capturing all relevant dynamics. read the letter →

arxiv 2606.01051 v1 pith:Y4RC7XBV submitted 2026-05-31 cs.LG

Xun Shen , Yuepeng Wang , Akifumi Wachi , Yongqi Zhou , Richard Weiss , Yoshihiko Fujisawa , Ken Kawano , Mehrshad Sadria
show 8 more authors
Ying Chen Xin Liu Sebastien Gros Xiao Hu Kyoung-Sook Kim Mengmou Li Katsuki Fujisawa Kenji Wakabayashi
This is my paper · ORCID
classification cs.LG
keywords continuous-timereinforcementlearningsafeRLmedicaltreatmentoptimizationsemi-Markovdecisionprocesstrajectorysafetyinteractiontimingpolicyfromlogs
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces a framework for safe continuous-time reinforcement learning in dynamic medical treatment, where decisions cover both treatment intensity and the timing of clinical visits amid continuously evolving patient states. It reformulates the problem as an option-based semi-Markov decision process in which each option encodes a treatment policy together with its duration. A safety-tightening mechanism is developed to show that discrete constraints applied at visit times extend to the entire trajectory between visits. Finite-sample guarantees are established for learning policies from logged data, and a practical data-driven conservative surrogate is provided for implementation. Experiments indicate that jointly optimizing visit timing yields higher safety and effectiveness than fixed equidistant schedules.

What carries the argument

The safety-tightening mechanism in the option-based semi-Markov decision process, where each option pairs a continuous-time treatment policy with a duration, transfers safety from discrete interaction points to the intervening trajectory.

What would settle it

Apply a learned policy in a continuous-time simulator containing adverse-event dynamics outside the semi-Markov model and check whether safety violations appear between interaction times despite satisfied discrete constraints.

Watch

Extended reading notes

Core claim

The central claim is that in an option-based semi-Markov decision process formulation of continuous-time medical treatment, suitably constructed constraints at interaction times guarantee safety over the full continuous-time trajectory with high probability, while finite-sample guarantees hold for policy learning from logged treatment trajectories and a data-driven conservative surrogate enables practical use.

Load-bearing premise

Continuous-time patient dynamics and adverse events between interactions are fully captured by the option-based semi-Markov formulation without unmodeled effects that break the safety transfer.

Editorial extensions

If this is right

  • Joint optimization of treatment policies and interaction timings is possible under trajectory-level safety constraints.
  • Finite-sample guarantees support policy learning directly from logged treatment trajectories.
  • A data-driven conservative surrogate provides a practical way to enforce the safety tightening.
  • Adaptive interaction timing improves both safety and treatment effectiveness compared with equidistant schemes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The approach could apply to other control settings with sparse observation opportunities and continuous underlying dynamics.
  • Fewer but better-timed clinical visits might become feasible while preserving safety margins.
  • Real-world deployment would need checks against patient dynamics that deviate from the semi-Markov assumption.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The paper proposes Interaction-Limited Safe Continuous-Time Reinforcement Learning for dynamical medical treatment. It reformulates the problem as an option-based semi-Markov decision process (SMDP) where options encode continuous-time treatment policies and durations, introduces a safety-tightening mechanism asserting that suitably constructed constraints at interaction times ensure safety over the full continuous-time trajectory with high probability, provides finite-sample guarantees for policy learning from logged data, and presents a data-driven conservative surrogate. Experiments are claimed to demonstrate that adaptive interaction timing improves safety and effectiveness over equidistant schemes across safe policy optimization methods.

Significance. If the safety transfer and finite-sample results hold under the stated modeling assumptions, the work would address a practically relevant gap in continuous-time safe RL for settings with limited clinical interactions, such as medical treatment scheduling. The explicit handling of interaction timing alongside treatment intensity, combined with trajectory-level guarantees, would be a substantive contribution over discrete-time or fixed-schedule baselines.

major comments (2)
  1. [Safety-tightening mechanism (abstract and theoretical development)] The central safety claim—that constraints at interaction times guarantee full-trajectory safety with high probability—rests on the option-based SMDP accurately capturing all continuous-time patient dynamics and adverse-event hazards between interactions. The manuscript provides no formal statement of this modeling assumption, no robustness analysis to unmodeled continuous-time effects, and no verification that the semi-Markov structure suffices (see skeptic note on the weakest assumption).
  2. [Finite-sample guarantees] Finite-sample guarantees for policy learning are asserted in the abstract, yet the provided text contains no derivation details, proof sketches, or verification steps, preventing assessment of whether the bounds are non-vacuous or depend on additional unstated assumptions.
minor comments (2)
  1. [Abstract] The abstract states that experiments show improvement but reports neither quantitative metrics, baselines, nor statistical significance; these details should be summarized with effect sizes.
  2. [Method] Notation for the option-based SMDP (e.g., option duration distributions, safety constraint tightening operator) should be introduced with explicit definitions before the safety mechanism is described.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comments point-by-point below, providing clarifications and committing to revisions that make the modeling assumptions and theoretical derivations explicit.

read point-by-point responses

  1. Referee: [Safety-tightening mechanism (abstract and theoretical development)] The central safety claim—that constraints at interaction times guarantee full-trajectory safety with high probability—rests on the option-based SMDP accurately capturing all continuous-time patient dynamics and adverse-event hazards between interactions. The manuscript provides no formal statement of this modeling assumption, no robustness analysis to unmodeled continuous-time effects, and no verification that the semi-Markov structure suffices (see skeptic note on the weakest assumption).

    Authors: We agree that the central modeling assumption requires an explicit formal statement. The assumption is that the continuous-time patient dynamics and adverse-event hazards between interactions are fully captured by the chosen option (treatment policy and duration) in the SMDP, with no unmodeled jumps or hazards outside this structure. We will add this as Assumption 3.1 in the revised manuscript. The safety-tightening mechanism then propagates the interaction-time constraints to the full trajectory via the option's continuous-time evolution. Regarding robustness, the high-probability guarantee is conditional on the model; we will add a dedicated limitations paragraph discussing potential misspecification. The semi-Markov structure is appropriate and sufficient because it directly encodes variable-duration continuous-time segments without requiring fixed discretization, which aligns with limited-interaction medical settings. The skeptic note referenced is not included in the report, but the structure holds under the standard Lipschitz and bounded-hazard conditions stated in Section 3. revision: yes

  2. Referee: [Finite-sample guarantees] Finite-sample guarantees for policy learning are asserted in the abstract, yet the provided text contains no derivation details, proof sketches, or verification steps, preventing assessment of whether the bounds are non-vacuous or depend on additional unstated assumptions.

    Authors: The finite-sample guarantees appear as Theorem 4.2, derived via concentration inequalities on the empirical conservative surrogate under the logged option trajectories. We acknowledge that the main text omitted the full derivation for space reasons. In the revision we will insert a concise proof sketch immediately after the theorem statement, outlining the application of vector-valued Hoeffding bounds to the SMDP value estimates, the dependence on trajectory count and option duration variance, and the role of the data-driven surrogate. This will make clear that the bounds are non-vacuous for moderate sample sizes and explicitly list all assumptions (e.g., bounded rewards, known option termination probabilities). revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; safety guarantee rests on external modeling assumptions

full rationale

The paper's key result (safety transfer from interaction-time constraints to full continuous-time trajectories) is derived from the option-based SMDP reformulation and a safety-tightening mechanism. No quoted equations or text reduce this claim to a self-definition, a fitted parameter renamed as prediction, or a load-bearing self-citation chain. The derivation is self-contained once the SMDP modeling assumption is granted; that assumption is stated as a premise rather than derived within the paper. No patterns matching the enumerated circularity kinds are exhibited.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

Review performed on abstract only; full derivations, assumptions, and experimental details unavailable for detailed ledger construction.

assumptions (1)
  • domain assumption Continuous-time treatment dynamics and adverse events can be represented via an option-based semi-Markov decision process.
    This is the central modeling choice enabling the reformulation and safety transfer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interaction-Limited Safe Continuous-Time RL for Dynamical Medical Treatment." pith.science (2026). https://pith.science/paper/Y4RC7XBV

@misc{pith2026260601051,
  author       = {Pith},
  title        = {Pith review of: Interaction-Limited Safe Continuous-Time RL for Dynamical Medical Treatment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y4RC7XBV}},
  note         = {Machine review of arXiv:2606.01051}
}
read the original abstract

Dynamic medical treatment requires deciding treatment intensity and intervention timing, while patient states evolve continuously and adverse events may occur between clinical interactions. Most existing treatment learning methods assume fixed schedules or enforce safety only at discrete decision points. We propose Interaction-Limited Safe Continuous-Time Reinforcement Learning, a framework that jointly optimizes treatment administration and clinical interaction timing under trajectory-level safety constraints. Our key idea is to reformulate the continuous time treatment problem as an option-based semi-Markov decision process, where each option specifies a continuous-time treatment policy and its duration. We develop a safety-tightening mechanism showing that suitably constructed constraints at interaction times guarantee safety over the full continuous-time trajectory with high probability. We further establish finite-sample guarantees for policy learning from logged treatment trajectories and introduce a practical data-driven conservative surrogate. Experiments show that the proposed adaptive interaction-timing mechanism improves both safety and treatment effectiveness over equidistant interaction schemes across different safe policy optimization methods.

Figures

Figures reproduced from arXiv: 2606.01051 by the authors.

Figure 1
Figure 1. Summary of the relations among the main theorems and constraints in Section 3. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Lactate evaluation results of different methods. The purple dashed horizontal line is the [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Comprehensive evaluation of different methods, averaged across five test seeds. (a) Mean [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Full-horizon PINN rollout versus observed clinical measurements for a representative ICU [PITH_FULL_IMAGE:figures/full_fig_p042_4.png]
Figure 5
Figure 5. Figure 5: Training process of CPO CPO-E remains at a higher cost level after the initial transient phase. Overall, the training curves suggest that adaptive interaction timing improves both treatment effectiveness and safety-related cost control for CPO at K = 12. Training with …
Figure 6
Figure 6. Figure 6: Training process of SAC [PITH_FULL_IMAGE:figures/full_fig_p045_6.png]
Figure 7
Figure 7. Figure 7: Training process of PCPO Both variants of PCPO show stable and well-behaved training trajectories. PCPO-E achieves faster early convergence, with more rapid reward improvement and quicker reduction in SOFA, while maintaining relatively smoother updates in the early sta…
Figure 8
Figure 8. Figure 8: Training process of TRPO J.4 State and Action Trajectories [PITH_FULL_IMAGE:figures/full_fig_p046_8.png]
Figure 9
Figure 9. Figure 9: Aggregated evaluation trajectories of CPO with adaptive (CPO-O) and equidistant (CPO-E) interaction timing at K = 12, averaged across five evaluation seeds. Panel (a) shows mean state trajectories, panel (b) shows mean treatment actions, and panel (c) highlights lactat…
Figure 10
Figure 10. Figure 10: Comprehensive evaluation of TRPO and PCPO across five test seeds. Panels (a) and (b) report mean SOFA scores per step under two evaluation settings, panel (c) shows the safety rate after t ≥ 20 hours across different interaction steps, and panel (d) shows representati…
Figure 11
Figure 11. Figure 11: Cost-limit sensitivity analysis for CPO and PCPO at K = 10. Panel (a) shows mean SOFA as the safety cost limit varies from 2.8 to 3.3, while panel (b) shows the post-20h lactate safety rate. Markers indicate mean values and shaded regions indicate variability across e…
Figure 12
Figure 12. Figure 12: Additional evaluation of PCPO and TRPO using a cluster-level PINN dynamics model. Panels (a) and (b) report the mean SOFA score per step under two evaluation summaries, panel (c) reports the post-20h lactate safety rate across interaction budgets, and panel (d) shows …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 4 canonical work pages

  1. [1]

    Constrained policy optimization

    Joshua Achiam, David Held, Aviv Tamar, and Pieter Abbeel. Constrained policy optimization. In Proceedings of the 34th International Conference on Machine Learning - Volume 70, ICML’17, page 22–31. JMLR.org, 2017

    2017

  2. [2]

    Routledge, 2021

    Eitan Altman.Constrained Markov decision processes. Routledge, 2021

    2021

  3. [3]

    The option-critic architecture

    Pierre-Luc Bacon, Jean Harb, and Doina Precup. The option-critic architecture. InProceedings of the AAAI conference on artificial intelligence, volume 31, 2017

    2017

  4. [4]

    Kellum, and Claudio Ronco

    Rinaldo Bellomo, John A. Kellum, and Claudio Ronco. Acute kidney injury.The Lancet, 380 (9843):756–766, 2012

    2012

  5. [5]

    R. S. Carbajo, J. Palma, and I. Martin-Loeches. The next frontier in sepsis: Connected icu data for real-world clinical decision making.Intensive Care Medicine, 52:301–308, 2026

    2026

  6. [6]

    A lyapunov-based approach to safe reinforcement learning.Advances in neural information processing systems, 31, 2018

    Yinlam Chow, Ofir Nachum, Edgar Duenez-Guzman, and Mohammad Ghavamzadeh. A lyapunov-based approach to safe reinforcement learning.Advances in neural information processing systems, 31, 2018

    2018

  7. [7]

    Serum lactate is associated with mortality in severe sepsis independent of organ failure and shock.Critical Care Medicine, 37(5):1670–1677, 2009

    Jason D Christie, Scarlett Bellamy, Chirag V Shah, Barry D Fuchs, Munish Goyal, David F Gaieski, Anne N Miltiades, and Mark E Mikkelsen. Serum lactate is associated with mortality in severe sepsis independent of organ failure and shock.Critical Care Medicine, 37(5):1670–1677, 2009

    2009

  8. [8]

    Bonferroni inequalities.The Annals of Probability, 5(4):577–581, 1977

    János Galambos. Bonferroni inequalities.The Annals of Probability, 5(4):577–581, 1977

    1977

Show all 45 references
  1. [9]

    Ghossein, B

    J. Ghossein, B. N. Hryciw, and K. Kyeremanteng. Redefining sepsis management: The comprehensive impact of artificial intelligence.Journal of Intensive Medicine, 5(2):134–136, September 2024

    2024

  2. [10]

    Guidelines for reinforcement learning in healthcare.Nature medicine, 25(1):16–18, 2019

    Omer Gottesman, Fredrik Johansson, Matthieu Komorowski, Aldo Faisal, David Sontag, Finale Doshi-Velez, and Leo Anthony Celi. Guidelines for reinforcement learning in healthcare.Nature medicine, 25(1):16–18, 2019

    2019

  3. [11]

    Grönwall

    Thomas H. Grönwall. Note on the derivatives with respect to a parameter of the solutions of a system of differential equations.Annals of Mathematics, 20(2):292–296, 1919

    1919

  4. [12]

    Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor

    Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. InProceedings of the 35th International Conference on Machine Learning, volume 80 ofProceedings of Machine Learnin...

    2018

  5. [13]

    Epicare: A reinforcement learning benchmark for dynamic treatment regimes

    Mason Hargrave, Alex Spaeth, and Logan Grosenick. Epicare: A reinforcement learning benchmark for dynamic treatment regimes. InAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  6. [14]

    S. Holt, A. Hüyük, and M. van der Schaar. Active observing in continuous-time control. In Advances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  7. [15]

    Reinforcement learning for sepsis treatment: A continuous action space solution

    Yong Huang, Rui Cao, and Amir Rahmani. Reinforcement learning for sepsis treatment: A continuous action space solution. InProceedings of the 7th Machine Learning for Healthcare Conference, volume 182 ofProceedings of Machine Learning Research, pages 631–647. PMLR, 2022

    2022

  8. [16]

    Hyndman, D

    Rob J. Hyndman, D. M. Bashtannyk, and G. K. Grunwald. Estimating and visualizing condi- tional densities.Journal of Computational and Graphical Statistics, 5:315–336, 1996

    1996

  9. [17]

    Optimal treatment strategies for critical patients with deep reinforcement learning.ACM Transactions on Intelligent Systems and Technology, 15(2):1–22, 2024

    Simi Job, Xiaohui Tao, Lin Li, Haoran Xie, Taotao Cai, Jianming Yong, and Qing Li. Optimal treatment strategies for critical patients with deep reinforcement learning.ACM Transactions on Intelligent Systems and Technology, 15(2):1–22, 2024. 11

    2024

  10. [18]

    Johnson, T.J

    A.E.W. Johnson, T.J. Pollard, L. Shen, L.H. Hehman, M. Feng, M.Ghassemi, B. Moody, P. Szolovits, L.A. Celi, and R.G. Mark. Mimic-iii, a freely accessible critical care database. 3 (may. 2016), 2016. URLhttps://doi.org/10.1038/sdata.2016.35

    2016 doi

  11. [19]

    Komorowski, L.A

    M. Komorowski, L.A. Celi, O. Badawi, A.C. Gordon, and A.A. Faisal. The artificial intelligence clinician learns optimal treatment strategies for sepsis in intensive care.Nature Medicine, 24: 1716–1720, 2018

    2018

  12. [20]

    Jedlicki, David Ouyang, Anthony Philippakis, David Sontag, and Ahmed M

    Keying Kuang, Frances Dean, Jack B. Jedlicki, David Ouyang, Anthony Philippakis, David Sontag, and Ahmed M. Alaa. Med-real2sim: Non-invasive medical digital twins using physics- informed self-supervised learning. InAdvances in Neural Information Processing Systems 37 (NeurIPS ...

    2024

  13. [21]

    Lingsch, Dana Grund, Siddhartha Mishra, and Georgios Kissas

    Levi E. Lingsch, Dana Grund, Siddhartha Mishra, and Georgios Kissas. Fuse: Fast unified simulation and estimation for pdes. InAdvances in Neural Information Processing Systems 37 (NeurIPS 2024), 2024

    2024

  14. [22]

    Time-regularized interrupting options (trio)

    Timothy Mann, Daniel Mankowitz, and Shie Mannor. Time-regularized interrupting options (trio). InProceedings of the 31st International Conference on Machine Learning, volume 32 of Proceedings of Machine Learning Research, pages 1350–1358. PMLR, 2014

    2014

  15. [23]

    Nambiar, S

    M. Nambiar, S. Ghosh, P. Ong, Y . E. Chan, Y . M. Bee, and P. Krishnaswamy. Deep offline reinforcement learning for real-world treatment optimization applications. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 4673–4684. ACM, 2023

    2023

  16. [24]

    Prescott, Massimo Antonelli, Waleed Alhazzani, Morten Hylander Møller, Fayez Alshamsi, Luciano C

    Hallie C. Prescott, Massimo Antonelli, Waleed Alhazzani, Morten Hylander Møller, Fayez Alshamsi, Luciano C. P. Azevedo, et al. Surviving sepsis campaign: International guidelines for management of sepsis and septic shock 2026.Critical Care Medicine, 54(4):725–812, 2026

    2026

  17. [25]

    Raissi, P

    M. Raissi, P. Perdikaris, and G.E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378:686–707, 2019. doi: https: //doi.org/10...

    2019 doi

  18. [26]

    Conservative continuous-time treatment optimization.arXiv preprint, 2026

    Nora Schneider, Georg Manten, and Niki Kilbertus. Conservative continuous-time treatment optimization.arXiv preprint, 2026. doi: 10.48550/arXiv.2603.16789

    2026 doi

  19. [27]

    Trust region policy optimization

    John Schulman, Sergey Levine, Pieter Abbeel, Michael Jordan, and Philipp Moritz. Trust region policy optimization. InProceedings of the 32nd International Conference on Machine Learning, volume 37 ofProceedings of Machine Learning Research, pages 1889–1897. PMLR, 07–09 Jul 2015

    2015

  20. [28]

    Probabilistic reachable sets of stochastic nonlinear systems with contextual uncertainties.Automatica, 176: 112237, 2025

    Xun Shen, Ye Wang, Kazumune Hashimoto, Yuhu Wu, and Sebastien Gros. Probabilistic reachable sets of stochastic nonlinear systems with contextual uncertainties.Automatica, 176: 112237, 2025

    2025

  21. [29]

    Deutschman, Christopher W

    Mervyn Singer, Clifford S. Deutschman, Christopher W. Seymour, Manu Shankar-Hari, Djillali Annane, Michael Bauer, Rinaldo Bellomo, Gordon R. Bernard, Jean-Daniel Chiche, Craig M. Coopersmith, et al. The third international consensus definitions for sepsis and septic shock (sep...

    2016

  22. [30]

    The spectrum of sepsis-associated encephalopathy: a clinical perspective.Critical Care, 27(1):386, 2023

    Romain Sonneville, Sarah Benghanem, Lina Jeantin, Etienne de Montmollin, Marc Doman, Augustin Gaudemer, Michael Thy, and Jean-François Timsit. The spectrum of sepsis-associated encephalopathy: a clinical perspective.Critical Care, 27(1):386, 2023

    2023

  23. [31]

    Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning.Artificial intelligence, 112(1-2): 181–211, 1999

    Richard S Sutton, Doina Precup, and Satinder Singh. Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning.Artificial intelligence, 112(1-2): 181–211, 1999

    1999

  24. [32]

    X. Tang, Y . Jia, J. Sun, and Y . Fan. Supervised reinforcement learning with recurrent neural network for dynamic treatment recommendation. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), pages 2447–2456. ACM, 2018. 12

    2018

  25. [33]

    Efficient exploration in continuous-time model-based reinforcement learning

    Lenart Treven, Jonas Hübotter, Bhavya Sukhija, Andreas Krause, and Florian Dörfler. Efficient exploration in continuous-time model-based reinforcement learning. InProceedings of the 37th International Conference on Neural Information Processing Systems, 2023

    2023

  26. [34]

    When to sense and control? a time-adaptive approach for continuous-time reinforcement learning

    Lenart Treven, Bhavya Sukhija, Yarden As, Florian Dörfler, and Andreas Krause. When to sense and control? a time-adaptive approach for continuous-time reinforcement learning. In Proceedings of the 38th International Conference on Neural Information Processing Systems (NeurIPS)...

  27. [35]

    Guardian-regularized safe offline reinforcement learning for smart weaning of mechanical circulatory devices

    Aysin Tumay, Sophia Sun, Sonia Fereidooni, Aaron Dumas, Elise Jortberg, and Rose Yu. Guardian-regularized safe offline reinforcement learning for smart weaning of mechanical circulatory devices. InProceedings of Machine Learning for Health (ML4H), volume 297 of Proceedings of ...

    2025

  28. [36]

    Cambridge University Press, 2018

    Roman Vershynin.High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge University Press, 2018

    2018

  29. [37]

    West and Andrew M

    John B. West and Andrew M. Luks.Respiratory Physiology: The Essentials. Lippincott Williams & Wilkins, 11th edition, 2020

    2020

  30. [38]

    Offline guarded safe reinforcement learning for medical treatment optimization strategies.Proceedings of the 39th Advances in Neural Information Processing Systems, 2025

    Runze Yan*, Xun Shen*, Akifumi Wachi, Sebastien Gros, Anni Zhao, and Xiao Hu. Offline guarded safe reinforcement learning for medical treatment optimization strategies.Proceedings of the 39th Advances in Neural Information Processing Systems, 2025

    2025

  31. [39]

    Tsung-Yen Yang, Justinian Rosca, Karthik Narasimhan, and Peter J. Ramadge. Projection-based constrained policy optimization. InInternational Conference on Learning Representations (ICLR), 2020

    2020

  32. [40]

    Yildiz, M

    C. Yildiz, M. Heinonen, and H. Lähdesmäki. Continuous-time model-based reinforcement learning. InProceedings of the 38th International Conference on Machine Learning (ICML), pages 12009–12018. PMLR, 2021

    2021

  33. [41]

    H. Yu, W. Xu, and H. Zhang. Taac: Temporally abstract actor-critic for continuous control. In Advances in Neural Information Processing Systems (NeurIPS), volume 34, pages 29021–29033, 2021

    2021

  34. [42]

    Continuous-time decision transformer for healthcare applications

    Zhiyue Zhang, Hongyuan Mei, and Yanxun Xu. Continuous-time decision transformer for healthcare applications. InProceedings of the 26th International Conference on Artificial Intelligence and Statistics (AISTATS), volume 206, pages 6245–6262. PMLR, 2023

    2023

  35. [43]

    estimated–ideal

    X. Zhou, Y . Jiao, J. Liu, and J. Huang. A deep generative approach to conditional sampling. Journal of the American Statistical Association, 118(543):1837–1848, 2023. doi: 10.1080/ 01621459.2021.2016424. 13 A Limitations This work has several limitations that suggest importan...

    2023

  36. [44]

    every feasible solution of(ECS)is a conservative surrogate ofh(x, δt)

  37. [45]

    35 Proof.SinceF(· |x, δt)is continuous and strictly increasing, the equation F(d s |x, δt) = 1−¯ϵ h has a unique solution

    with probability at least 1−ρ h, the approximate solution ˜vNpr,Nre ¯ϵh (x, δt) is a conservative surrogate ofh(x, δt). 35 Proof.SinceF(· |x, δt)is continuous and strictly increasing, the equation F(d s |x, δt) = 1−¯ϵ h has a unique solution. Hence the minimization problem (EC...

Pith tools

Reviewed June 28, 2026 · model on record in the stance chip above.