Pith. sign in

REVIEW 3 major objections 5 minor 27 references

PRECISE-AS: Personalized Reinforcement Learning for Efficient Point-of-Care Echocardiography in Aortic Stenosis Diagnosis

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read RL-driven active video acquisition matches full echocardiography accuracy while using 47% of clips

desk verdict The 47% video-reduction claim lacks a random-subset baseline; since the classifier is trained with 50% token masking, random selection of two videos may already match full acquisition, so the RL agent's contribution is unverified. read the letter →

arxiv 2509.02898 v1 pith:LQ7PM5M5 submitted 2025-09-02 cs.CV

Armin Saadat , Nima Hashemi , Hooman Vaseli , Michael Y. Tsang , Christina Luong , Michiel Van de Panne , Teresa S. M. Tsang , Purang Abolmaesumi This is my paper · ORCID
classification cs.CV
keywords aorticstenosisechocardiographypoint-of-careultrasoundactivefeatureacquisitionreinforcementlearningvideoclassificationDoubleDQNclinicaldecisionsupport
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 tries to show that a reinforcement-learning agent can decide, patient by patient, which echocardiography video clips to keep and when to stop acquiring more, without hurting diagnostic accuracy for aortic stenosis. On 2,572 patient studies, the system reaches 80.6% balanced accuracy for three severity classes while using 47% of the clips a full study would use—about 1.88 of the four clips considered—and this matches both its own full-acquisition classifier and the state-of-the-art ProtoASNet baseline with no statistically significant difference. The authors argue this is the first active-video-acquisition framework for echocardiography and that it could make point-of-care ultrasound faster and more accessible by replacing fixed protocols with personalized stopping decisions. A reader should care because the result is a concrete demonstration that sequential, cost-aware image selection can be applied to high-dimensional medical video, not just tabular data or single static images.

What carries the argument

The load-bearing object is the active video acquisition MDP: state s is an N×D matrix holding the feature embeddings of selected videos with zero vectors for unselected positions; actions are acquire video i or terminate; and the only nonzero reward, R(sT)=1[correct]−λΣci, is given at termination. Double DQN learns the policy over this MDP using features from the frozen ProtoASNet encoder, while a Transformer with positional encodings and a CLS token performs study-level classification and supplies the correctness term. The design makes the problem a personalized stopping problem: when the expected diagnostic value of another clip no longer exceeds its cost, the agent terminates.

What would settle it

Give the trained policy a real-time acquisition setting where candidate clips are not already present: an operator uses the policy's chosen view label to steer the probe, and the metric is total scan time or number of clips acquired until the policy terminates. If matching the 47% clip reduction requires acquiring more than 1.88 clips to obtain the selected ones (for example, because the first PLAX clip is unusable and must be retaken), the claimed efficiency gain fails to transfer. Alternatively, a prospective multicenter study could check whether the RL-selected subset yields 80.6% balanced

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that echocardiography video acquisition for aortic stenosis severity assessment can be modeled as a Markov decision process and solved with reinforcement learning to produce per-patient acquisition policies. Starting from a zeroed state over four fixed, ordered clips (two parasternal long-axis, two parasternal short-axis), the agent either selects one more clip or terminates; the terminal state is classified by a Transformer that treats selected clips as tokens and a missing clip as a masked token. The reward is +1 for a correct severity label minus a cost coefficient times the number of clips acquired. Trained with Double DQN on frozen features

Load-bearing premise

The efficiency result rests on modeling a study as exactly four predetermined clips—two PLAX and two PSAX chosen in advance, each with unit cost—so the 47% reduction is a saving over four clips, not over the number of video clips a sonographer would acquire in a live point-of-care exam.

Editorial extensions

If this is right

  • At the λ=0.001 operating point, the policy uses 47% of the four-clip acquisition (1.88 clips per study) at 80.6% balanced accuracy; the same classifier without RL needs all four clips to reach that accuracy.
  • The cost coefficient λ gives a tunable efficiency-accuracy frontier: λ=0.01 yields 41% of clips at 79.9% balanced accuracy, and λ=0.25 yields 31% at 77.1%, so a deployment could pick an operating point before scanning.
  • The learned acquisition order—typically a PLAX clip first, then a PSAX clip—is interpretable and consistent with complementary views of the aortic valve, suggesting the policy can expose underlying acquisition rules.
  • Because the policy matches the full-acquisition accuracy within statistical noise (p>0.05), the framework's practical value is efficiency at equal accuracy rather than improved accuracy.

Reading between the lines

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

  • The 47% figure counts savings over four pre-selected clips, not over live scanning time; if a sonographer must acquire and process more than the chosen clips to find good representatives of each view, the realized savings will be smaller than 47%.
  • One testable extension is to let the policy request a view label rather than a specific clip index and combine it with a real-time view detector, turning the system into a probe-guidance assistant that tells the operator which standard view to capture next.
  • The recipe—frozen feature encoder, masked-token Transformer classifier, and RL over acquisition actions—should transfer to other multi-view medical imaging or sequential diagnostic tests, where the same accuracy-versus-cost tradeoff exists.
  • Since the agent learns to prefer first PLAX then PSAX, a simpler non-RL protocol of fixed view ordering might achieve part of the savings; comparing RL against that fixed-order baseline would isolate the value of learning personalized stopping.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes PRECISE-AS, an RL-driven active video acquisition framework for aortic stenosis (AS) severity classification from echocardiography. It encodes each of four pre-selected echo clips (PLAX1, PLAX2, PSAX1, PSAX2) with ProtoASNet, trains a Transformer classifier under 50% random token masking, and learns a DDQN policy that sequentially acquires or skips clips and terminates when confident. On a private dataset of 2,572 patients (252 test), the method reports 80.6% balanced accuracy while acquiring on average 47% of the videos (1.88 of 4 clips), matching the full-acquisition version (80.6%) and the ProtoASNet baseline (80.0%); the authors disclose that the improvement over ProtoASNet is not statistically significant (p > 0.05). The paper includes an ablation over the cost coefficient lambda and qualitative acquisition-pathway analysis.

Significance. If the efficiency claim survives scrutiny, the paper is a plausible contribution to active feature acquisition in high-dimensional video data, an under-explored area in medical imaging. The strengths are concrete: a large private clinical dataset with patient-level splits, held-out test set, repeated runs with standard deviations, honest reporting of the non-significant accuracy gap versus ProtoASNet, and public code. The central methodological idea—learning a personalized acquisition policy over frozen video-token embeddings—is reasonable. However, the load-bearing efficiency claim (47% video reduction at equal accuracy) is not yet internally validated, because no random-subset baseline is reported despite the classifier being explicitly trained for 50% missing tokens. The significance of the paper therefore depends on the outcome of that missing control experiment.

major comments (3)
  1. [§3.3, §4.3, Table 1] The central efficiency claim is not supported without a random-subset baseline. The classifier is deliberately trained with 50% random token masking (§3.3), so it is robust to missing approximately two of four videos by construction. If PRECISE-AS (with RL) acquires 1.88 videos on average, a trivial random or fixed subset of two videos (e.g., [PLAX1, PSAX1]) may achieve the same 80.6% bACC simply because the classifier was trained for the missing-token regime. Table 1 compares against 100%-video baselines only; Table 2 sweeps λ but never includes a same-cost random policy. The observed 47% figure could therefore be an artifact of the masking rate rather than of the learned selection policy. Please add random-k and fixed-subset curves across k=1..4, and report the performance gap between RL and random selection at matched average acquisition counts.
  2. [§4.1] The deployment claim is more limited than the abstract suggests. The MDP operates over exactly four pre-selected clips per patient ordered [PLAX1, PLAX2, PSAX1, PSAX2], with unit cost per clip. The 47% is therefore a reduction from 4 predetermined clips to 1.88 on average, not a reduction in the number of images a sonographer would capture in a real POCUS exam. View selection, clip count, variable clip quality, and operator time are not modeled. At a minimum, the title/abstract should say "among four pre-selected echo clips," and the discussion should acknowledge this simplification as a boundary on the efficiency claim.
  3. [§3.3, Table 2] The 50% masking rate and the cost coefficient λ are free parameters that jointly determine the efficiency result. λ=0.001 is selected post hoc; no principled selection criterion or sensitivity analysis is given, and the claim of "using only 47%" is one point on the λ sweep. Similarly, the masking rate is fixed at 50% without evidence that this rate matches the RL agent's typical acquisition policy. Please report performance over a range of masking rates, and define how λ is chosen in practice (e.g., by a validation-based accuracy/cost trade-off) before claiming the 47% operating point as the headline result.
minor comments (5)
  1. [Eq. (1)] The notation m_i · f(x_i) with m_i a boolean scalar is unconventional; as written this is scalar multiplication, not a dot product. Replace with m_i f(x_i) or clarify.
  2. [§3.2, Eq. (2)] The update s_{t+1}[i] = f(x_i) is dimensionally inconsistent if s is an N×D state; it should read s_{t+1}[i, :] = f(x_i) or otherwise state that the i-th row is replaced.
  3. [§4.2] DDQN hyperparameters are under-specified: no replay buffer size, ε-greedy schedule, target network update frequency, learning rate, or batch size. This limits reproducibility despite the code release.
  4. [§4.1] Because two PLAX and two PSAX videos are "randomly selected" per patient, the four-clip set is not deterministic. Please state whether all patients had exactly two usable clips per view, and report variance over the random clip selection; otherwise the 47% average count may itself be a random quantity.
  5. [Fig. 2] The left panel is described as F1 score versus average number of acquired videos, but the surrounding text sometimes refers to accuracy. Clarify the metric and the meaning of the shaded regions (standard deviation over runs, presumably).

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the efficiency claim is an empirical measurement on held-out patients; the main concern is a missing random-subset baseline, which is a validity issue rather than circularity.

full rationale

The paper makes no first-principles derivation that reduces to its inputs. PRECISE-AS is an active feature acquisition system: a ProtoASNet encoder and Transformer classifier are trained and frozen, then a DDQN agent is trained to select videos using a reward that depends on the frozen classifier's predictions. The reported 80.6% bACC at 47% acquired videos is a measured operating point on the held-out test set, not a quantity derived from the training data by construction. The λ sweep (Table 2) selects the cost coefficient that yields the headline result; this is hyperparameter selection, not fitting a parameter and renaming it a prediction. The self-citation of ProtoASNet as the feature extractor is a dependency, not load-bearing evidence: ProtoASNet is an open-source, independently trained model, and its use does not preordain the RL agent's video selections or the test-set accuracy. The more substantive concern is that the classifier is trained with 50% random masking (Section 3.3), which may make it robust to missing videos; without a random-subset baseline, the efficiency gain could be partly attributable to this masking robustness rather than to the RL policy. However, this is a missing control and an internal-validity issue, not circular reasoning: the RL agent is not equivalent to its inputs by definition, and no equation or construction forces the reported result. Thus the paper earns a low non-circularity score with no specific circular steps.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central result rests on the frozen ProtoASNet features, the four-clip-per-study construction, the 50% masking rate, and the cost coefficient lambda. No new physical entities are introduced. The most consequential choices are the artificial four-clip MDP and the lambda-dependent operating point.

free parameters (2)
  • cost coefficient lambda = 0.001 (used for headline result)
    Trades classification accuracy against video cost in Eq. 3; Table 2 sweeps values, and the paper's reported 47% point is at lambda=0.001.
  • token masking rate during classifier training = 0.5
    Chosen to make the classifier robust to missing videos; no tuning or sensitivity analysis is shown for this rate.
assumptions (3)
  • domain assumption Frozen ProtoASNet features f(xi) are a sufficient representation for both classification and acquisition decisions.
    Section 3.1 uses ProtoASNet as the fixed encoder; if the features omit clinically relevant information, both the classifier and the RL policy are limited by them.
  • domain assumption A study can be represented by exactly four clips in a fixed view order [PLAX1, PLAX2, PSAX1, PSAX2] with unit video cost.
    Section 4.1 randomly selects two PLAX and two PSAX videos per patient and orders them; the MDP's state and action space are built on this N=4 structure.
  • domain assumption The RL policy trained on rewards from the frozen classifier generalizes across patients and does not select subsets that exploit classifier artifacts.
    Section 3.2 trains DDQN with terminal reward based on the frozen classifier; the evaluation assumes this policy transfers to held-out patients.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PRECISE-AS: Personalized Reinforcement Learning for Efficient Point-of-Care Echocardiography in Aortic Stenosis Diagnosis." pith.science (2026). https://pith.science/paper/LQ7PM5M5

@misc{pith2026250902898,
  author       = {Pith},
  title        = {Pith review of: PRECISE-AS: Personalized Reinforcement Learning for Efficient Point-of-Care Echocardiography in Aortic Stenosis Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LQ7PM5M5}},
  note         = {Machine review of arXiv:2509.02898}
}
read the original abstract

Aortic stenosis (AS) is a life-threatening condition caused by a narrowing of the aortic valve, leading to impaired blood flow. Despite its high prevalence, access to echocardiography (echo), the gold-standard diagnostic tool, is often limited due to resource constraints, particularly in rural and underserved areas. Point-of-care ultrasound (POCUS) offers a more accessible alternative but is restricted by operator expertise and the challenge of selecting the most relevant imaging views. To address this, we propose a reinforcement learning (RL)-driven active video acquisition framework that dynamically selects each patient's most informative echo videos. Unlike traditional methods that rely on a fixed set of videos, our approach continuously evaluates whether additional imaging is needed, optimizing both accuracy and efficiency. Tested on data from 2,572 patients, our method achieves 80.6% classification accuracy while using only 47% of the echo videos compared to a full acquisition. These results demonstrate the potential of active feature acquisition to enhance AS diagnosis, making echocardiographic assessments more efficient, scalable, and personalized. Our source code is available at: https://github.com/Armin-Saadat/PRECISE-AS.

Figures

Figures reproduced from arXiv: 2509.02898 by the authors.

Figure 1
Figure 1. Overview of the structure. Each action ai acquires the i-th video and updates the state. At termination (aT ), the final state sT is passed on for classification. moving to a new state given the current state and action, while the reward function R : S × A × S → R assigns rewards based on these transitions. An episode is a sequence of states, actions, and rewards from an initial state to a terminal state. A policy π… view at source ↗
Figure 2
Figure 2. (Left) F1 score versus the average number of acquired videos per study. Each yellow data point corresponds to a specific cost-coefficient. The solid lines represent the mean performance, while the shaded regions indicate one standard deviation. (Right) A sample study consisting of four echo videos. PRECISE-AS only requires the top two videos to achieve the same patient-level accuracy as the whole study. lines. Video… view at source ↗
Figure 3
Figure 3. Personalized diagnostic pathways for AS classification. Each node represents a state defined by the set of acquired videos and shows how many patients reached it. Directed edges illustrate the sequential video acquisitions leading to a final diagnosis [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 21 canonical work pages

  1. [1]

    IEEE Transactions on Medical Imaging 43(1), 366–376 (2024)

    Ahmadi, N., Tsang, M.Y., Gu, A.N., Tsang, T.S.M., Abolmaesumi, P.: Transformer-based spatio-temporal analysis for classification of aortic stenosis severity from echocardiography cine series. IEEE Transactions on Medical Imaging 43(1), 366–376 (2024). https://doi.org/10.1109/TMI.2023.3305384

    work page arXiv 2024

  2. [2]

    Improving Skin Condition Classification with a Question Answering Model

    Akrout, M., Farahmand, A., Jarmain, T.: Improving skin condition classification with a question answering model. CoRR abs/1811.06165 (2018). https://doi. org/10.48550/arXiv.1811.06165

    work page Pith review arXiv doi:10.48550/arxiv.1811.06165 2018

  3. [3]

    In: Medical Image Computing and Computer Assisted Intervention – MICCAI

    Akrout, M., Farahmand, A., Jarmain, T., Abid, L.: Improving skin condition clas- sification with a visual symptom checker trained using reinforcement learning. In: Medical Image Computing and Computer Assisted Intervention – MICCAI

    work page

  4. [4]

    In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2022

    Bernardino,G.,Jonsson,A.,Loncaric,F.,MartíCastellote,P.,Sitges,M.,Clarysse, P., Duchateau, N.: Reinforcement learning for active modality selection during diagnosis. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2022. LNCS, vol. 13431, pp. 592–601. Springer (2022)

    work page 2022

  5. [5]

    Carabello, B.A.: Introduction to aortic stenosis. Circ. Res.113(2), 179–185 (2013). https://doi.org/10.1161/CIRCRESAHA.113.300156

    work page doi:10.1161/circresaha.113.300156 2013

  6. [6]

    8928–8939 (2019)

    Chen, C., Li, O., Tao, C., Barnett, A.J., Rudin, C., Su, J.K.: This looks like that: deep learning for interpretable image recognition pp. 8928–8939 (2019). https: //doi.org/10.48550/arXiv.1806.10574

    arXiv doi:10.48550/arxiv.1806.10574 2019

  7. [7]

    Proceedings of Machine Learning Research,vol

    Chen, Y., Hassani, S.H., Karbasi, A., Krause, A.: Sequential information maxi- mization: when is greedy near-optimal? In: Proceedings of the 28th Conference on Learning Theory (COLT 2015). Proceedings of Machine Learning Research,vol. 40, pp. 338–363. PMLR (2015), https://proceedings.mlr.press/v40/Chen15b.html

    work page 2015

  8. [8]

    Learning to Maximize Mutual Information for Dynamic Feature Selection

    Covert, I.C., Qiu, W., Lu, M., Kim, N.Y., White, N.J., Lee, S.: Learning to maximize mutual information for dynamic feature selection. In: Proceedings of the 40th International Conference on Machine Learning (ICML 2023). Proceed- ings of Machine Learning Research, vol. 202, pp. 6424–6447. PMLR (2023). https://doi.org/10.48550/arXiv.2301.00557 10 A. Saadat et al

    work page Pith review arXiv doi:10.48550/arxiv.2301.00557 2023

Show all 27 references
  1. [9]

    In: Simplifying Med- ical Ultrasound – ASMUS 2021

    Ginsberg, T., Tal, R., Tsang, M.Y., Macdonald, C., Taheri Dezaki, F., van der Kuur, J., Luong, C., Abolmaesumi, P., Tsang, T.S.M.: Deep video networks for automatic assessment of aortic stenosis in echocardiography. In: Simplifying Med- ical Ultrasound – ASMUS 2021. LNCS, vol....

    2021

  2. [10]

    In: DataPerf: Benchmarking Data for Data-Centric AI Workshop at ICML 2022 (2022)

    Huang, Z., Long, G., Wessler, B., Hughes, M.C.: TMED 2: a dataset for semi- supervised classification of echocardiograms. In: DataPerf: Benchmarking Data for Data-Centric AI Workshop at ICML 2022 (2022)

    2022

  3. [11]

    15714–15723 (2021)

    Kim, E., Kim, S., Seo, M., Yoon, S.: XProtoNet: diagnosis in chest radiography with global and local explanations pp. 15714–15723 (2021). https://doi.org/10. 1109/CVPR46437.2021.01546

    2021

  4. [12]

    CoRR abs/2107.04163 (2021)

    Li, Y., Shan, S., Liu, Q., Oliva, J.B.: Towards robust active feature acquisition. CoRR abs/2107.04163 (2021). https://doi.org/10.48550/arXiv.2107.04163

    2021 doi arXiv

  5. [13]

    IEEE Transactions on Medical Imaging39(6), 1868–1883 (2020)

    Liao, Z., Girgis, H., Abdi, A., Vaseli, H., Hetherington, J., Rohling, R., Gin, K., Tsang, T., Abolmaesumi, P.: On modelling label uncertainty in deep neural networks: automatic estimation of intra observer variability in 2D echocardiogra- phy quality assessment. IEEE Transact...

    2020

  6. [14]

    In: International Encyclopedia of the Social & Behavioral Sciences, pp

    Littman, M.L.: Markov decision processes. In: International Encyclopedia of the Social & Behavioral Sciences, pp. 9240–9242. Elsevier, Oxford (2001)

    2001

  7. [15]

    Luong, C.L., Ong, K., Kaila, K., Pellikka, P.A., Gin, K., Tsang, T.S.M.: Focused cardiac ultrasonography: current applications and future directions. J. Ultrasound Med. 38(4), 865–876 (2019). https://doi.org/10.1002/jum.14773

    2019 doi

  8. [16]

    Munt,B.,O’Neill,B.J.,Koilpillai,C.,Gin,K.,Jue,J.,Honos,G.:Treatingtheright patient at the right time: access to echocardiography in canada. Can. J. Cardiol. 22(12), 1029–1033 (2006). https://doi.org/10.1016/S0828-282X(06)70318-X

    2006 doi

  9. [17]

    Muyama, L., Neuraz, A., Coulet, A.: Deep reinforcement learning for personal- ized diagnostic decision pathways using electronic health records: a comparative study on anemia and systemic lupus erythematosus. Artif. Intell. Med.157, 102994 (2024). https://doi.org/10.1016/j.art...

    2024

  10. [18]

    Sanfilippo, A.J., Bewick, D., Chan, K.L., Cujec, B., Dumesnil, J.G., Honos, G., etal.:Guidelinesfortheprovisionofechocardiographyincanada:recommendations of a joint canadian cardiovascular society/canadian society of echocardiography consensus panel. Can. J. Cardiol. 21(9), 76...

    2005

  11. [19]

    Strange, G., Stewart, S., Celermajer, D., Prior, D., Scalia, G.M., Marwick, T., Ilton, M., Joseph, M., Codde, J., Playford, D.: Poor long-term survival in patients with moderate aortic stenosis. J. Am. Coll. Cardiol. 74(15), 1851–1863 (2019). https://doi.org/10.1016/j.jacc.2019.08.004

    2019 doi

  12. [20]

    In: Proceedings of the Thirtieth AAAI Conference on Artificial Intel- ligence (AAAI-16)

    Van Hasselt, H., Guez, A., Silver, D.: Deep reinforcement learning with double Q-learning. In: Proceedings of the Thirtieth AAAI Conference on Artificial Intel- ligence (AAAI-16). pp. 2094–2100. AAAI Press (2016). https://doi.org/10.48550/ arXiv.1509.06461

    doi arXiv

  13. [21]

    In: Medical Image Computing and Computer Assisted In- tervention – MICCAI 2023

    Vaseli, H., Gu, A.N., Ahmadi Amiri, S.N., Tsang, M.Y., Fung, A., Kondori, N., Saadat, A., Abolmaesumi, P., Tsang, T.S.M.: ProtoASNet: dynamic prototypes for inherently interpretable and uncertainty-aware aortic stenosis classification in echocardiography. In: Medical Image Com...

    2023 doi

  14. [22]

    https://doi.org/ 10.48550/arXiv.1706.03762 Title Suppressed Due to Excessive Length 11

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need30, 5998–6008 (2017). https://doi.org/ 10.48550/arXiv.1706.03762 Title Suppressed Due to Excessive Length 11

    2017 doi arXiv

  15. [23]

    Watkins, C.J.C.H., Dayan, P.: Q-learning. Mach. Learn.8(3–4), 279–292 (1992). https://doi.org/10.1007/BF00992698

    1992 doi

  16. [24]

    In: NeurIPS 2020 Workshop on Machine Learning for Health (ML4H) (2020)

    Yin, H., Li, Y., Pan, S.J., Zhang, C., Tschiatschek, S.: Reinforcement learning with efficient active feature acquisition. In: NeurIPS 2020 Workshop on Machine Learning for Health (ML4H) (2020). https://doi.org/10.48550/arXiv.2011.00825

    2020 doi arXiv

  17. [25]

    In: Proceedings of the Eleventh In- ternational Conference on Learning Representations (ICLR 2023)

    Yu, Z., Li, Y., Kim, J.C., Huang, K., Luo, Y., Wang, M.: Deep reinforcement learning for cost-effective medical diagnosis. In: Proceedings of the Eleventh In- ternational Conference on Learning Representations (ICLR 2023). OpenReview (2023). https://doi.org/10.48550/arXiv.2302.10261

    2023 doi arXiv

  18. [26]

    In: ICML 2019 Workshop on Real-world Sequential Decision Making

    Zannone, S., Hernández-Lobato, J.M., Zhang, C., Palla, K.: Odin: optimal discov- ery of high-value information using model-based deep reinforcement learning. In: ICML 2019 Workshop on Real-world Sequential Decision Making. pp. 1–8 (2019), https://realworld-sdm.github.io/paper/21.pdf

    2019

  19. [2019]

    11767, pp

    LNCS, vol. 11767, pp. 549–557. Springer (2019). https://doi.org/10.1007/ 978-3-030-32251-9_60

    2019

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.