REVIEW 3 major objections 1 minor 54 references
Toward Optimal Statistical Inference in Noisy Linear Quadratic Reinforcement Learning over a Finite Horizon
T0 review · 3 major / 1 minor · reviewed 2026-08-05 · deepseek-v4-flash
Pith's one-line read Noisy LQ reinforcement learning gets valid, online confidence intervals
desk verdict The paper is unverifiable from the supplied text — corrupted mojibake with the wrong arXiv header — but the abstract describes a plausible and important contribution that deserves a clean-copy review. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the online bootstrap for the policy-gradient estimate: maintain the PG update and a collection of randomly perturbed copies of it, and refresh all of them with each new observation. The load-bearing identity is the exact asymptotic distribution of the PG estimator (and of the objective loss), which lets the perturbed copies reproduce the estimator's sampling distribution. The $n^{-1/4}$ quantile-approximation rate is what allows finite-sample confidence statements.
What would settle it
Run the proposed online bootstrap on a scalar LQ system while driving the exploration-noise variance to zero at a fast rate; if the resulting confidence intervals under-cover as n grows (coverage below nominal by more than simulation noise), the excitation condition is violated and the asymptotic claim fails. Alternatively, simulate with a heavy-tailed disturbance (e.g., t-distributed with 3 degrees of freedom) and check whether the empirical coverage of the bootstrap intervals falls below the nominal level.
Extended reading notes
Core claim
The central claim is that uncertainty quantification for policy-gradient methods in noisy LQ RL is tractable. For linear dynamics with quadratic cost over a finite horizon, the paper derives exact asymptotic distributions for both the policy-gradient estimator and the corresponding objective loss. It then proposes an online bootstrap procedure: alongside the main PG estimate, maintain a set of randomly perturbed PG estimates, updating all of them as new observations arrive. The paper proves this procedure is distributionally consistent, meaning the bootstrap distribution tracks the estimator's sampling distribution, and that confidence intervals built from it have asymptotic and non-asymptot
Load-bearing premise
The load-bearing premise is that the noise and exploration are well behaved: the dynamics disturbance is zero-mean with sufficiently light tails and enough moments, and the exploration noise keeps the state covariance non-degenerate over the finite horizon, so the drift parameter and policy-gradient estimator are identifiable, asymptotically normal, and have an invertible covariance.
Editorial extensions
If this is right
- Learned LQ policies can be reported with confidence intervals, not just point estimates, in both offline and streaming settings.
- The $n^{-1/4}$ quantile error gives a concrete sample-size target for a desired interval precision.
- The result covers unknown drift parameters, so it applies when the system model is partially identified from data.
- It provides a template for uncertainty quantification in other differentiable RL objectives: plug in the estimator, add perturbed copies, and update them online.
- In high-stakes control, operators can flag policies whose expected-cost intervals are too wide before deployment.
Reading between the lines
- The proof likely rests on a fixed-horizon martingale or CLT argument; extending to discounted infinite-horizon LQ may need a different mixing argument, since the asymptotic covariance changes.
- The $n^{-1/4}$ rate is characteristic of bootstrap quantile accuracy under a CLT; a natural test is whether the coverage error in simulations decays as $n^{-1/4}$ rather than $n^{-1/2}$.
- Heavy-tailed disturbance or weak exploration noise would break the normal asymptotics; a robust bootstrap (e.g., based on self-normalization) would be a natural follow-up.
- The method should transfer to LQ regulators with stochastic cost or partial-state observations, but the exact covariance would need recomputation.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper claims to establish statistical inference foundations for policy gradient (PG) methods in finite-horizon noisy Linear Quadratic Reinforcement Learning (LQ RL), for both known and unknown drift parameters. The abstract states that exact asymptotic distributions are derived for the PG estimator and the objective loss, and that an online bootstrap procedure, which updates perturbed PG estimates alongside the main estimate, is distributionally consistent and yields confidence intervals with both asymptotic and non-asymptotic validity. A specific quantile approximation rate of n^{-1/4} is claimed, with applicability to offline and fully online settings and numerical experiments. The only readable content is the abstract; the supplied full text is unreadable mojibake on every page, and the running headers identify the document as arXiv:2508.08441v3 [q-bio.QM], not arXiv:2508.08436. Consequently, no theorem statement, assumption, proof, algorithm description, or numerical result could be inspected.
Significance. If the results hold, the paper would address an important gap: uncertainty quantification for policy-gradient methods in LQ RL, an area where most existing work focuses on convergence rates rather than distributional inference. Exact asymptotics for PG estimators and an online bootstrap with a non-asymptotic n^{-1/4} quantile error would be a substantive theoretical contribution with clear practical relevance for high-stakes sequential decision-making. However, the evidentiary basis is currently absent: no mathematical claim can be verified from the provided manuscript. The paper cannot be credited for machine-checked proofs or reproducible code, because none are accessible. The potential significance is real, but the manuscript in its current form does not permit any technical assessment.
major comments (3)
- [Full text (all pages)] The supplied full text is unreadable mojibake on every page, and the page headers identify the document as arXiv:2508.08441v3 [q-bio.QM] (9 May 2026), not arXiv:2508.08436. No theorem statement, assumption list, proof, or experimental detail can be audited. This blocks every central claim in the abstract: exact asymptotic distributions, distributional consistency of the online bootstrap, asymptotic and non-asymptotic validity, and the n^{-1/4} quantile approximation rate. This is not a cosmetic defect; it is a load-bearing evidentiary problem that makes any substantive verification impossible.
- [Abstract] The abstract does not state the regularity conditions needed to support the claims: zero-mean and moment assumptions on the dynamics disturbance, sufficient excitation from the exploration noise, identifiability of the drift parameter, and invertibility of the asymptotic covariance of the PG estimator. These conditions are exactly what would determine whether the claimed n^{-1/4} rate follows rather than a slower Berry-Esseen-type bound, and whether the bootstrap is valid in the fully online setting. Because the assumptions section is unreadable, the central claims rest on unverifiable premises.
- [Numerical experiments] The abstract mentions numerical experiments across noisy linear dynamical systems, but no figures, tables, or readable text from the experiments can be inspected. Thus there is no way to assess the empirical support for the claimed finite-sample behavior or the practical performance of the proposed intervals.
minor comments (1)
- [Submission file] The uploaded PDF must be replaced with a legible version whose running headers match the arXiv identifier and subject class of this submission; the current rendering is unusable.
Circularity Check
No circularity identifiable in readable material; the supplied full text is unreadable mojibake from a different arXiv ID, so the body cannot be audited.
full rationale
The only legible portion is the abstract, which presents theorem-shaped claims: exact asymptotic distributions for the PG estimators and objective loss, distributional consistency of the online bootstrap, asymptotic and non-asymptotic validity, and an n^{-1/4} quantile-approximation rate. These are stated as derived outputs rather than as fitted inputs, and no fitted parameter is relabeled as a prediction in the readable text. There is also no visible self-citation invoked as load-bearing, no imported uniqueness theorem forcing a choice, and no ansatz smuggled in via citation. The supplied full text is corrupted mojibake whose page headers identify arXiv:2508.08441v3 [q-bio.QM], not arXiv:2508.08436; consequently no equation-level reduction (Eq. X = Eq. Y by construction) can be quoted or exhibited. Under the rule that circularity may be claimed only when the paper's own text permits exhibiting the specific reduction, no circular step can be scored from the available evidence. The corrupted rendering and the arXiv-ID mismatch are evidentiary and verifiability concerns, not a demonstration of circularity.
Assumptions & free parameters
assumptions (4)
- domain assumption Linear dynamics with additive zero-mean noise and quadratic cost; the optimal policy is linear and the PG objective has the standard LQ structure.
- domain assumption Regularity of noise and exploration: moments, sufficient excitation, identifiable drift, invertible asymptotic covariance.
- standard math Standard stochastic approximation and bootstrap consistency machinery: suitable step sizes, Lipschitz gradient, and smoothness of the objective so the online bootstrap distribution converges to the same limit as the PG estimator.
- domain assumption Boundedness over the finite horizon: costs and state covariances do not blow up, so the objective and its gradient are well behaved.
Cite this review
Pith. "Pith review of Toward Optimal Statistical Inference in Noisy Linear Quadratic Reinforcement Learning over a Finite Horizon." pith.science (2026). https://pith.science/paper/CLDORF7D
@misc{pith2026250808436,
author = {Pith},
title = {Pith review of: Toward Optimal Statistical Inference in Noisy Linear Quadratic Reinforcement Learning over a Finite Horizon},
year = {2026},
howpublished = {\url{https://pith.science/paper/CLDORF7D}},
note = {Machine review of arXiv:2508.08436}
}
abstract
Recent developments in Reinforcement learning have significantly enhanced sequential decision-making in uncertain environments. Despite their strong performance guarantees, most existing work has focused primarily on improving the operational accuracy of learned control policies and the convergence rates of learning algorithms, with comparatively little attention to uncertainty quantification and statistical inference. Yet, these aspects are essential for assessing the reliability and variability of control policies, especially in high-stakes applications. In this paper, we study statistical inference for the policy gradient (PG) method for noisy Linear Quadratic Reinforcement learning (LQ RL) over a finite time horizon, where linear dynamics with both known and unknown drift parameters are controlled subject to a quadratic cost. We establish the theoretical foundations for statistical inference in LQ RL, deriving exact asymptotics for both the PG estimators and the corresponding objective loss. Furthermore, we introduce a principled inference framework that leverages online bootstrapping to construct confidence intervals for both the learned optimal policy and the corresponding objective losses. The method updates the PG estimates along with a set of randomly perturbed PG estimates as new observations arrive. We prove that the proposed bootstrapping procedure is distributionally consistent and that the resulting confidence intervals achieve both asymptotic and non-asymptotic validity. Notably, our results imply that the quantiles of the exact distribution can be approximated at a rate of $n^{-1/4}$, where $n$ is the number of samples used during the procedure. The proposed procedure is easy to implement and applicable to both offline and fully online settings. Numerical experiments illustrate the effectiveness of our approach across a range of noisy linear dynamical systems.
Reference graph
Works this paper leans on
-
[1]
Normal approximation for stochastic gradient descent via non-asymptotic rates of martingale clt
Andreas Anastasiou, Krishnakumar Balasubramanian, and Murat A Erdogdu. Normal approximation for stochastic gradient descent via non-asymptotic rates of martingale clt. In Conference on Learning Theory , pages 115--137. PMLR, 2019
work page 2019
-
[2]
Karl Johan str \"o m. Adaptive control. In Mathematical System Theory: The Influence of RE Kalman , pages 437--450. Springer, 1995
work page 1995
-
[3]
Matteo Basei, Xin Guo, Anran Hu, and Yufei Zhang. Logarithmic regret for episodic continuous-time linear-quadratic reinforcement learning over a finite-time horizon. Journal of Machine Learning Research , 23(178):1--34, 2022
work page 2022
-
[4]
Adaptive control with the stochastic approximation algorithm: Geometry and convergence
Arthur Becker, P Kumar, and Ching-Zong Wei. Adaptive control with the stochastic approximation algorithm: Geometry and convergence. IEEE Transactions on Automatic Control , 30(4):330--338, 1985
work page 1985
-
[5]
Dynamic programming and optimal control: Volume I , volume 4
Dimitri Bertsekas. Dynamic programming and optimal control: Volume I , volume 4. Athena scientific, 2012
work page 2012
-
[6]
Reinforcement learning applied to linear quadratic regulation
Steven Bradtke. Reinforcement learning applied to linear quadratic regulation. Advances in Neural Information Processing Systems , 5, 1992
work page 1992
-
[7]
Linear-quadratic mean-field reinforcement learning: convergence of policy gradient methods
Ren \'e Carmona, Mathieu Lauri \`e re, and Zongjun Tan. Linear-quadratic mean-field reinforcement learning: convergence of policy gradient methods. arXiv preprint arXiv:1910.04295 , 2019
arXiv 1910
-
[8]
Statistical inference for online decision making via stochastic gradient descent
Haoyu Chen, Wenbin Lu, and Rui Song. Statistical inference for online decision making via stochastic gradient descent. Journal of the American Statistical Association , 116(534):708--719, 2021
work page 2021
Show all 54 references
-
[9]
Statistical inference for model parameters in stochastic gradient descent
Xi Chen, Jason D Lee, Xin T Tong, and Yichen Zhang. Statistical inference for model parameters in stochastic gradient descent. 2020
2020
-
[10]
Robust inference via multiplier bootstrap
Xi Chen and Wen-Xin Zhou. Robust inference via multiplier bootstrap. 2020
2020
-
[11]
On the sample complexity of the linear quadratic regulator
Sarah Dean, Horia Mania, Nikolai Matni, Benjamin Recht, and Stephen Tu. On the sample complexity of the linear quadratic regulator. Foundations of Computational Mathematics , 20(4):633--679, 2020
2020
-
[12]
Challenges of reinforcement learning
Zihan Ding and Hao Dong. Challenges of reinforcement learning. Deep Reinforcement Learning: Fundamentals, Research and Applications , pages 249--272, 2020
2020
-
[13]
Challenges of real-world reinforcement learning: definitions, benchmarks and analysis
Gabriel Dulac-Arnold, Nir Levine, Daniel J Mankowitz, Jerry Li, Cosmin Paduraru, Sven Gowal, and Todd Hester. Challenges of real-world reinforcement learning: definitions, benchmarks and analysis. Machine Learning , 110(9):2419--2468, 2021
2021
-
[14]
On the convergence theory of debiased model-agnostic meta-reinforcement learning
Alireza Fallah, Kristian Georgiev, Aryan Mokhtari, and Asuman Ozdaglar. On the convergence theory of debiased model-agnostic meta-reinforcement learning. Advances in Neural Information Processing Systems , 34:3096--3107, 2021
2021
-
[15]
Online bootstrap confidence intervals for the stochastic gradient descent estimator
Yixin Fang, Jinfeng Xu, and Lei Yang. Online bootstrap confidence intervals for the stochastic gradient descent estimator. Journal of Machine Learning Research , 19(78):1--21, 2018
2018
-
[16]
Global convergence of policy gradient methods for the linear quadratic regulator
Maryam Fazel, Rong Ge, Sham Kakade, and Mehran Mesbahi. Global convergence of policy gradient methods for the linear quadratic regulator. In International conference on machine learning , pages 1467--1476. PMLR, 2018
2018
-
[17]
Convergence of policy gradient methods for finite-horizon exploratory linear-quadratic control problems
Michael Giegrich, Christoph Reisinger, and Yufei Zhang. Convergence of policy gradient methods for finite-horizon exploratory linear-quadratic control problems. SIAM Journal on Control and Optimization , 62(2):1060--1092, 2024
2024
-
[18]
Reinforcement learning for linear-convex models with jumps via stability analysis of feedback controls
Xin Guo, Anran Hu, and Yufei Zhang. Reinforcement learning for linear-convex models with jumps via stability analysis of feedback controls. SIAM Journal on Control and Optimization , 61(2):755--787, 2023
2023
-
[19]
Policy gradient methods for the noisy linear quadratic regulator over a finite horizon
Ben Hambly, Renyuan Xu, and Huining Yang. Policy gradient methods for the noisy linear quadratic regulator over a finite horizon. SIAM Journal on Control and Optimization , 59(5):3359--3391, 2021
2021
-
[20]
Bootstrapping upper confidence bound
Botao Hao, Yasin Abbasi Yadkori, Zheng Wen, and Guang Cheng. Bootstrapping upper confidence bound. Advances in Neural Information Processing Systems , 32, 2019
2019
-
[21]
Model-based or model-free, a review of approaches in reinforcement learning
Qingyan Huang. Model-based or model-free, a review of approaches in reinforcement learning. In 2020 International Conference on Computing and Data Science (CDS) , pages 219--221. IEEE, 2020
2020
-
[22]
Improved zeroth-order variance reduced algorithms and analysis for nonconvex optimization
Kaiyi Ji, Zhe Wang, Yi Zhou, and Yingbin Liang. Improved zeroth-order variance reduced algorithms and analysis for nonconvex optimization. In International conference on machine learning , pages 3100--3109. PMLR, 2019
2019
-
[23]
On incomplete learning and certainty-equivalence control
N Bora Keskin and Assaf Zeevi. On incomplete learning and certainty-equivalence control. Operations Research , 66(4):1136--1167, 2018
2018
-
[24]
Iterated least squares in multiperiod control
Tze Leung Lai and Herbert Robbins. Iterated least squares in multiperiod control. Advances in Applied Mathematics , 3(1):50--73, 1982
1982
-
[25]
Fast inference for quantile regression with tens of millions of observations
Sokbae Lee, Yuan Liao, Myung Hwan Seo, and Youngki Shin. Fast inference for quantile regression with tens of millions of observations. Journal of Econometrics , page 105673, 2024
2024
-
[26]
Statistical estimation and inference via local sgd in federated learning
Xiang Li, Jiadong Liang, Xiangyu Chang, and Zhihua Zhang. Statistical estimation and inference via local sgd in federated learning. arXiv preprint arXiv:2109.01326 , 2021
2021 arXiv
-
[27]
Unifying offline causal inference and online bandit learning for data driven decision
Ye Li, Hong Xie, Yishi Lin, and John CS Lui. Unifying offline causal inference and online bandit learning for data driven decision. In Proceedings of the Web Conference 2021 , pages 2291--2303, 2021
2021
-
[28]
A generalized reinforcement-learning model: Convergence and applications
Michael L Littman and Csaba Szepesv \'a ri. A generalized reinforcement-learning model: Convergence and applications. In ICML , volume 96, pages 310--318, 1996
1996
-
[29]
A review of uncertainty for deep reinforcement learning
Owen Lockwood and Mei Si. A review of uncertainty for deep reinforcement learning. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment , volume 18, pages 155--162, 2022
2022
-
[30]
Gradient estimation in model-based reinforcement learning: a study on linear quadratic environments
\^A ngelo Greg \'o rio Lovatto, Thiago Pereira Bueno, and Leliane Nunes de Barros. Gradient estimation in model-based reinforcement learning: a study on linear quadratic environments. In Brazilian Conference on Intelligent Systems , pages 33--47. Springer, 2021
2021
-
[31]
Certainty equivalence is efficient for linear quadratic control
Horia Mania, Stephen Tu, and Benjamin Recht. Certainty equivalence is efficient for linear quadratic control. Advances in Neural Information Processing Systems , 32, 2019
2019
-
[32]
A linear quadratic regulator based speed control for remote-controlled racing cars
Olusesi Ayobami Meadows, Arthur Rodriguez, and Ahmed Tijani Salawudeen. A linear quadratic regulator based speed control for remote-controlled racing cars. In 2022 IEEE Nigeria 4th International Conference on Disruptive Technologies for Sustainable Development (NIGERCON) , pag...
2022
-
[33]
Black-box generalization: Stability of zeroth-order learning
Konstantinos Nikolakakis, Farzin Haddadpour, Dionysis Kalogerias, and Amin Karbasi. Black-box generalization: Stability of zeroth-order learning. Advances in neural information processing systems , 35:31525--31541, 2022
2022
-
[34]
Acceleration of stochastic approximation by averaging
Boris T Polyak and Anatoli B Juditsky. Acceleration of stochastic approximation by averaging. SIAM Journal on Control and Optimization , 30(4):838--855, 1992
1992
-
[35]
Temporal difference models: Model-free deep rl for model-based control
Vitchyr Pong, Shixiang Gu, Murtaza Dalal, and Sergey Levine. Temporal difference models: Model-free deep rl for model-based control. arXiv preprint arXiv:1802.09081 , 2018
2018 arXiv
-
[36]
Online bootstrap inference for policy evaluation in reinforcement learning
Pratik Ramprasad, Yuantong Li, Zhuoran Yang, Zhaoran Wang, Will Wei Sun, and Guang Cheng. Online bootstrap inference for policy evaluation in reinforcement learning. Journal of the American Statistical Association , 118(544):2901--2914, 2023
2023
-
[37]
The unintended consequences of discount regularization: Improving regularization in certainty equivalence reinforcement learning
Sarah Rathnam, Sonali Parbhoo, Weiwei Pan, Susan Murphy, and Finale Doshi-Velez. The unintended consequences of discount regularization: Improving regularization in certainty equivalence reinforcement learning. In International Conference on Machine Learning , pages 28746--287...
2023
-
[38]
Implicit bias of policy gradient in linear quadratic control: Extrapolation to unseen initial states
Noam Razin, Yotam Alexander, Edo Cohen-Karlik, Raja Giryes, Amir Globerson, and Nadav Cohen. Implicit bias of policy gradient in linear quadratic control: Extrapolation to unseen initial states. arXiv preprint arXiv:2402.07875 , 2024
2024 arXiv
-
[39]
Efficient estimations from a slowly convergent robbins-monro process
David Ruppert. Efficient estimations from a slowly convergent robbins-monro process. Technical report, Cornell University Operations Research and Industrial Engineering, 1988
1988
-
[40]
Gaussian approximation and multiplier bootstrap for polyak-ruppert averaged linear stochastic approximation with applications to td learning
Sergey Samsonov, Eric Moulines, Qi-Man Shao, Zhuo-Song Zhang, and Alexey Naumov. Gaussian approximation and multiplier bootstrap for polyak-ruppert averaged linear stochastic approximation with applications to td learning. In The Thirty-eighth Annual Conference on Neural Infor...
2024
-
[41]
Dynamic causal effects evaluation in a/b testing with a reinforcement learning framework
Chengchun Shi, Xiaoyu Wang, Shikai Luo, Hongtu Zhu, Jieping Ye, and Rui Song. Dynamic causal effects evaluation in a/b testing with a reinforcement learning framework. Journal of the American Statistical Association , 118(543):2059--2071, 2023
-
[42]
Learning optimal controllers by policy gradient: Global optimality via convex parameterization
Yue Sun and Maryam Fazel. Learning optimal controllers by policy gradient: Global optimality via convex parameterization. In 2021 60th IEEE Conference on Decision and Control (CDC) , pages 4576--4581. IEEE, 2021
2021
-
[43]
Optimal scheduling of entropy regularizer for continuous-time linear-quadratic reinforcement learning
Lukasz Szpruch, Tanut Treetanthiploet, and Yufei Zhang. Optimal scheduling of entropy regularizer for continuous-time linear-quadratic reinforcement learning. SIAM Journal on Control and Optimization , 62(1):135--166, 2024
2024
-
[44]
Robust exploration in linear quadratic reinforcement learning
Jack Umenberger, Mina Ferizbegovic, Thomas B Sch \"o n, and H kan Hjalmarsson. Robust exploration in linear quadratic reinforcement learning. Advances in Neural Information Processing Systems , 32, 2019
2019
-
[45]
Distributed lqr design for identical dynamically coupled systems: Application to load frequency control of multi-area power grid
Eleftherios E Vlahakis, Leonidas D Dritsas, and George D Halikias. Distributed lqr design for identical dynamically coupled systems: Application to load frequency control of multi-area power grid. In 2019 IEEE 58th Conference on Decision and Control (CDC) , pages 4471--4476. I...
2019
-
[46]
First-order regret in reinforcement learning with linear function approximation: A robust estimation approach
Andrew J Wagenmaker, Yifang Chen, Max Simchowitz, Simon Du, and Kevin Jamieson. First-order regret in reinforcement learning with linear function approximation: A robust estimation approach. In International Conference on Machine Learning , pages 22384--22429. PMLR, 2022
2022
-
[47]
Residual bootstrap exploration for bandit algorithms
Chi-Hua Wang, Yang Yu, Botao Hao, and Guang Cheng. Residual bootstrap exploration for bandit algorithms. arXiv preprint arXiv:2002.08436 , 2020
2002 arXiv
-
[48]
Exact asymptotics for linear quadratic adaptive control
Feicheng Wang and Lucas Janson. Exact asymptotics for linear quadratic adaptive control. Journal of Machine Learning Research , 22(265):1--112, 2021
2021
-
[49]
Continuous-time mean--variance portfolio selection: A reinforcement learning framework
Haoran Wang and Xun Yu Zhou. Continuous-time mean--variance portfolio selection: A reinforcement learning framework. Mathematical Finance , 30(4):1273--1308, 2020
2020
-
[50]
Stochastic zeroth-order optimization in high dimensions
Yining Wang, Simon Du, Sivaraman Balakrishnan, and Aarti Singh. Stochastic zeroth-order optimization in high dimensions. In International conference on artificial intelligence and statistics , pages 1356--1365. PMLR, 2018
2018
-
[51]
Leveraging linear quadratic regulator cost and energy consumption for ultrareliable and low-latency iot control systems
Haojun Yang, Kuan Zhang, Kan Zheng, and Yi Qian. Leveraging linear quadratic regulator cost and energy consumption for ultrareliable and low-latency iot control systems. IEEE Internet of Things Journal , 7(9):8356--8371, 2020
2020
-
[52]
Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost
Zhuoran Yang, Yongxin Chen, Mingyi Hong, and Zhaoran Wang. Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost. Advances in Neural Information Processing Systems , 32, 2019
2019
-
[53]
Online covariance matrix estimation in stochastic gradient descent
Wanrong Zhu, Xi Chen, and Wei Biao Wu. Online covariance matrix estimation in stochastic gradient descent. Journal of the American Statistical Association , 118(541):393--404, 2023
2023
-
[54]
Uncertainty quantification and exploration for reinforcement learning
Yi Zhu, Jing Dong, and Henry Lam. Uncertainty quantification and exploration for reinforcement learning. Operations Research , 72(4):1689--1709, 2024
2024
Reviewed August 5, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.