Pith. sign in

REVIEW 1 major objections 19 references

Extending Causal Metamodeling to a non-Markovian Queue

T0 review · 1 major / 0 minor · reviewed 2026-06-28 · grok-4.3

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

Pith's one-line read Approximating general distributions with phase-type distributions extends modular dynamic Bayesian networks to non-Markovian queues.

desk verdict The paper takes a first cut at causal metamodeling for non-Markovian queues by swapping in phase-type approximations, but the validation on how well that approximation holds up is missing. read the letter →

arxiv 2606.00795 v1 pith:WNJONC2O submitted 2026-05-30 cs.LG cs.AI

Pracheta Amaranath , Anant Bhide , David Jensen , Peter Haas This is my paper · ORCID
classification cs.LGcs.AI
keywords causalmetamodelingmodulardynamicBayesiannetworksnon-Markovianqueuesphase-typedistributionsG/M/1queueprobabilisticqueriesdiscrete-eventsimulation
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

This paper extends modular dynamic Bayesian networks, which previously worked only for Markovian systems, to non-Markovian queues. It achieves the extension by replacing non-exponential distributions with phase-type approximations. The work identifies and gives initial solutions for the resulting issues of phase count selection, parameter learning, and sampling interval choice. Experiments on a G/M/1 queue confirm that the extended model answers probabilistic and causal queries accurately while delivering large speedups over direct simulation.

What carries the argument

Modular dynamic Bayesian networks (MDBNs) extended by phase-type distribution approximations to represent non-Markovian continuous-time dynamics in a discrete-time model.

What would settle it

Running the MDBN and direct simulation in parallel on the same G/M/1 queue and finding that the MDBN answers to probabilistic causal queries deviate substantially from the simulation results.

Watch

Extended reading notes

Core claim

By approximating non-exponential distributions using phase-type distributions, modular dynamic Bayesian networks can be extended to non-Markovian queues, yielding the first causal metamodeling technique for such systems. Experiments on a G/M/1 queue demonstrate that the MDBN can produce accurate answers to PCQs with orders-of-magnitude speedup of inference times relative to direct simulation.

Load-bearing premise

Approximating non-exponential distributions with phase-type distributions preserves enough accuracy for the metamodel to answer probabilistic and causal queries correctly.

Editorial extensions

If this is right

  • A single trained model estimates a range of probabilistic and causal queries on the non-Markovian system.
  • Inference times are orders of magnitude shorter than those of direct simulation.
  • The number of phases can be chosen to balance metamodel accuracy against computational cost.
  • The same framework applies to other non-Markovian queueing models once the phase-type approximation is set.

Reading between the lines

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

  • The speedup may allow real-time causal analysis in applied queueing settings such as service systems or networks.
  • Similar phase-type approximations could be tested on multi-server or network-of-queues models.
  • Automatic methods for selecting the number of phases might be developed to reduce manual tuning.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

1 major / 0 minor

Summary. The paper extends modular dynamic Bayesian networks (MDBNs) for causal metamodeling from Markovian to non-Markovian queueing systems. It approximates general (non-exponential) distributions via phase-type distributions to enable estimation of probabilistic and causal queries (PCQs) from a single trained model. The work identifies three challenges—selecting the number of phases to balance accuracy and tractability, efficient parameter learning, and choosing a sampling interval for discrete-time approximation of continuous-time dynamics—and offers preliminary solutions. Experiments on a G/M/1 queue are reported to show that the resulting MDBN yields accurate PCQ answers while providing orders-of-magnitude speedup over direct simulation.

Significance. If the central claims hold, this would constitute the first causal metamodeling approach applicable to non-Markovian systems, addressing a clear limitation of prior MDBN work. The ability to answer a range of PCQs from one model without repeated expensive simulations could be valuable for queueing analysis in operations research and performance modeling, where general distributions are the norm.

major comments (1)
  1. [Experiments section (G/M/1 queue results)] Experiments section (G/M/1 queue results): the manuscript reports that the MDBN produces 'accurate answers' to PCQs but contains no sensitivity analysis, error tables, or quantitative results showing how PCQ error varies with the number of phases used in the phase-type approximation. Because the abstract itself identifies the accuracy-tractability tradeoff as a novel challenge and the central claim rests on this approximation, the absence of such validation leaves the accuracy assertion unsupported.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thoughtful review and for highlighting the need for stronger validation of the phase-type approximation. We address the major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Experiments section (G/M/1 queue results)] Experiments section (G/M/1 queue results): the manuscript reports that the MDBN produces 'accurate answers' to PCQs but contains no sensitivity analysis, error tables, or quantitative results showing how PCQ error varies with the number of phases used in the phase-type approximation. Because the abstract itself identifies the accuracy-tractability tradeoff as a novel challenge and the central claim rests on this approximation, the absence of such validation leaves the accuracy assertion unsupported.

    Authors: We agree that the current experiments section provides only a qualitative demonstration of accuracy on the G/M/1 queue without quantitative sensitivity analysis or error tables versus the number of phases. The manuscript is framed as initiating the extension with preliminary solutions to the identified challenges, which is why the validation was kept at a high level. To strengthen the central claim, we will add in the revised version a new subsection (or expanded experiments) that includes (i) error tables for representative PCQs as a function of phase count (e.g., 2, 5, 10 phases) and (ii) a plot or table showing the accuracy-tractability tradeoff. This will directly address the referee's concern while preserving the preliminary nature of the work. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected

full rationale

The paper extends prior MDBN work (cited as external) to non-Markovian queues via phase-type distribution approximations, introduces new challenges and preliminary solutions for phase count, parameter learning, and sampling interval, then validates via G/M/1 experiments showing PCQ accuracy and speedup. No derivation step reduces by construction to its inputs, no fitted parameter is relabeled as a prediction, and no load-bearing claim rests solely on self-citation chains; the core extension and empirical results remain independent of the target quantities.

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

Abstract-only review provides no identifiable free parameters, axioms, or invented entities; all such elements remain unknown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Extending Causal Metamodeling to a non-Markovian Queue." pith.science (2026). https://pith.science/paper/WNJONC2O

@misc{pith2026260600795,
  author       = {Pith},
  title        = {Pith review of: Extending Causal Metamodeling to a non-Markovian Queue},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNJONC2O}},
  note         = {Machine review of arXiv:2606.00795}
}
read the original abstract

Metamodels for discrete-event simulations approximate the behavior of simulation models without running expensive simulations. Prior work introduced modular dynamic Bayesian networks (MDBNs) -- a class of metamodels that can estimate a range of probabilistic and causal queries (PCQs) using a single, trained model -- but the method was limited to Markovian systems. In this paper, we initiate an extension of MDBNs to non-Markovian queues by approximating non-exponential distributions using phase-type distributions. This approach raises novel challenges, including balancing metamodeling accuracy and tractability when choosing the number of phases, efficiently learning metamodel parameters, and choosing the sampling interval that is used to approximate a continuous-time simulation by a discrete-time MDBN. We provide preliminary solutions to these challenges, yielding the first causal metamodeling technique for non-Markovian systems. Experiments on a G/M/1 queue demonstrate that the MDBN can produce accurate answers to PCQs with orders-of-magnitude speedup of inference times relative to direct simulation.

Figures

Figures reproduced from arXiv: 2606.00795 by the authors.

Figure 1
Figure 1. MDBN model for the M/M/1 queue For the M/M/1 queue studied in prior work, the MDBN consists of three variables: arrival rate λ, service rate µ, and queue length L. The structure of the MDBN is depicted in Figure 1a—the input parameters λ and µ are parents of L in every time slice, i.e., each discrete time step at which the system state is recorded. The queue length at each time slice (j +1) depends on the queue leng… view at source ↗
Figure 2
Figure 2. Phase-type (GED) approximations of target distributions for the G/M/1 queue. Each plot shows the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. MDBN model for the G/M/1 queue 3.4 Optimal Sampling Interval In prior work that modeled the M/M/1 queue with MDBNs, the sampling interval δ was selected empirically by evaluating PCQ accuracy across a range of input parameter values. This was feasible because the M/M/1 queue has only two input parameters (λ,µ), making the computation tractable. As we move toward G/M/1 (and potentially G/G/1) queues, the number of in… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: (left) CPD parameters estimated using MLE—note the lack of training data for [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Evaluation of the MDBN on the Gamma/M/1 queue. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Evaluation of the MDBN on the Weibull/M/1 queue. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Evaluation of the MDBN on the Beta/M/1 queue. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 1 canonical work pages

  1. [1]

    2009 , publisher=

    Causality , author=. 2009 , publisher=

    2009

  2. [2]

    1994 , publisher=

    Matrix-Geometric Solutions in Stochastic Models: An Algorithmic Approach , author=. 1994 , publisher=

    1994

  3. [3]

    2023 Winter Simulation Conference (WSC) , pages=

    Causal Dynamic Bayesian Networks for Simulation Metamodeling , author=. 2023 Winter Simulation Conference (WSC) , pages=. 2023 , organization=

    2023

  4. [4]

    Scandinavian Journal of Statistics , pages=

    Fitting phase-type distributions via the EM algorithm , author=. Scandinavian Journal of Statistics , pages=. 1996 , publisher=

    1996

  5. [5]

    Proceedings of the 2020 Winter Simulation Conference , pages=

    Tutorial: Metamodeling for Simulation , author=. Proceedings of the 2020 Winter Simulation Conference , pages=. 2020 , editor =

    2020

  6. [6]

    Operations-Research-Spektrum , volume=

    On the use of phase type distributions in reliability modelling of systems with two components , author=. Operations-Research-Spektrum , volume=. 1981 , publisher=

    1981

  7. [7]

    IMA Journal of Applied Mathematics , volume=

    Numerical methods and hypoexponential approximations for gamma distributed delay differential equations , author=. IMA Journal of Applied Mathematics , volume=. 2022 , publisher=

    2022

  8. [8]

    IFAC Proceedings Volumes , volume=

    Approximation of continuous distribution via the generalized erlang distribution , author=. IFAC Proceedings Volumes , volume=. 2009 , publisher=

    2009

Show all 19 references
  1. [9]

    Pi Mu Epsilon Journal , pages=

    Time discretization of Markov chains , author=. Pi Mu Epsilon Journal , pages=. 2010 , publisher=

    2010

  2. [10]

    Artificial Intelligence , volume=

    Lazy Propagation: A Junction Tree Inference Algorithm based on Lazy Evaluation , author=. Artificial Intelligence , volume=. 1999 , publisher=

    1999

  3. [11]

    Ducamp, Gaspard and Gonzales, Christophe and Wuillemin, Pierre-Henri , BOOKTITLE =

  4. [12]

    Journal of the American statistical association , volume=

    Statistical analysis of a telephone call center: A queueing-science perspective , author=. Journal of the American statistical association , volume=. 2005 , publisher=

    2005

  5. [13]

    Lifetime data analysis , volume=

    A Weibull regression model with gamma frailties for multivariate survival data , author=. Lifetime data analysis , volume=. 1997 , publisher=

    1997

  6. [14]

    Computers & Operations Research , volume=

    Queueing-model based analysis of assembly lines with finite buffers and general service times , author=. Computers & Operations Research , volume=. 2008 , publisher=

    2008

  7. [15]

    arXiv preprint arXiv:1301.7362 , year=

    Tractable inference for complex stochastic processes , author=. arXiv preprint arXiv:1301.7362 , year=

    arXiv

  8. [16]

    2009 , address =

    Probabilistic Graphical Models: Principles and Techniques , author=. 2009 , address =

    2009

  9. [17]

    1992 , publisher=

    Regenerative Stochastic Simulation , author=. 1992 , publisher=

    1992

  10. [18]

    Management Science , volume=

    Better estimation of PERT activity time parameters , author=. Management Science , volume=. 1993 , publisher=

    1993

  11. [19]

    write newline

    " write newline "" before.all 'output.state := FUNCTION article output.bibitem format.authors "author" output.check author format.key output output.year.check new.block format.title "title" output.check new.block crossref missing format.jour.vol output format.article.crossref ...

Pith tools

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