Pith. sign in

REVIEW 5 major objections 6 minor 58 references

Complex Model Transformations by Reinforcement Learning with Uncertain Human Guidance

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper argues that human advice about beneficial states, even when the advisor is uncertain, can be quantified and fused into a reinforcement learning agent's policy to substantially speed up learning of complex model transformation…

desk verdict Worth a look for MDE+RL researchers, but the paper's own t-tests undermine its headline claim that uncertain advice substantially improves RL performance. read the letter →

arxiv 2506.20883 v2 pith:3RR74V76 submitted 2025-06-25 cs.SE cs.AIcs.LG

Kyanna Dagenais , Istvan David This is my paper · ORCID
classification cs.SEcs.AIcs.LG
keywords modeltransformationreinforcementlearninghumanguidancesubjectivelogicpolicyshapinguncertaintymodel-drivenengineeringopinionfusion
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

Writing long chains of model transformations by hand is error-prone, and pure reinforcement learning is slow to converge at the start of training. This paper argues that human advice, even when the human is not confident, can steer an RL agent to good transformation sequences much faster. It presents a framework that maps user-defined model transformations onto RL primitives, converts advice about beneficial states into subjective-logic opinions with explicit uncertainty, and fuses those opinions into the agent's policy. In a scaled Frozen Lake evaluation, advised agents outperformed the unadvised baseline in nearly every configuration, and a human advisor with only 10% coverage matched an oracle at high uncertainty. If the result transfers to real model-driven engineering tasks, domain experts could accelerate model synchronization, repair, and design-space exploration with sparse, imperfect tips rather than exact instructions.

What carries the argument

The load-bearing machinery is the opinion-fusion pipeline. Advice $\alpha\colon s\mapsto v$ is compiled into a binomial subjective-logic opinion $\omega=(b,d,u,a)$, with base rate $a=1/|A|$, uncertainty $u$ calibrated from a distance metric (linear discount, optionally with a threshold $\tau$), and belief and disbelief split by the advice value. The agent's probability-valued policy is translated into the certainty domain, the opinion for an advised state is fused with the opinions of the policy entries in its neighborhood (the states from which it is reachable, per the paper's Definitions 1 and 2) using the Belief Constraint Fusion operator, and the fused policy is normalized back into probabilities. This is the step that converts sparse, imperfect human tips into a changed exploration strategy before the agent acts.

What would settle it

Run the same advised-versus-unadvised comparison on a real model synchronization or repair problem whose states are graph morphisms rather than grid cells; if human advice at 60% or higher uncertainty does not raise cumulative reward above the unadvised baseline, the central claim fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that uncertainty in human guidance can be modeled and exploited rather than treated as a nuisance. A human opinion about whether a state is beneficial is compiled into a binomial subjective-logic opinion $\omega=(b,d,u,a)$, propagated to the neighboring states from which the advised state is reachable, and fused into the agent's policy with the Belief Constraint Fusion operator. The shaped policy is then converted back into probabilities and used to prioritize model transformations. In 30 runs on a $12\times 12$ Frozen Lake map with 20% holes, agents guided by oracle or single-human advice at uncertainty levels $u=0.0$ through $u=0.8$ accumulated more cumulative reward than the unadvised baseline in all but one configuration, and at moderate-to-high uncertainty a low-quota human advisor performed comparably to a full-knowledge oracle. The paper reads this as evidence that early, imperfect advice can be more valuable than late certain advice.

Load-bearing premise

The approach rests on the assumption that blending a person's advice about one spot into the agent's action choices for adjacent spots guides learning in the intended direction, and this was tested only on a small grid world, not on real model-transformation problems.

Editorial extensions

If this is right

  • A domain expert can express advice in a small domain-specific language by naming a state and a benefit value; the framework turns those statements into automatic policy changes.
  • Advice with uncertainty up to $u=0.4$ reliably beats the unadvised agent, and at $u\geq 0.6$ a human advisor with a 10% advice quota does not systematically lose to an oracle with full information.
  • Because the chosen fusion operator is commutative, any number of advisors can contribute without additional machinery; the paper notes that cooperating humans also outperform the unadvised baseline.
  • The learned policy can be exploited two ways: as a runtime conflict resolver that prioritizes among matched model transformations, or as a standalone generated transformation chain used independently of the RL engine.
  • When advisors are certain, giving more advice pays off; when they are uncertain, sparse advice is the better strategy.

Reading between the lines

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

  • Editorial extension: the evaluation defines neighboring states topologically, as adjacent grid cells. The method's usefulness in real model-driven engineering depends on whether a sensible neighborhood can be defined on graph morphisms, where one transformation application away is the natural notion; the paper does not test that.
  • Editorial extension: the uncertainty calibration scheme, a distance-based linear discount, is one of many. Inter-annotator agreement among several experts could also be turned into an uncertainty value and fed through the same fusion pipeline, which might fit multi-view modeling settings better than physical distance.
  • Editorial extension: the paper's timeliness argument suggests a concrete design rule not directly measured here: give advisors a small early budget of advice rather than asking for complete coverage, since low quotas matched the oracle at high uncertainty in the reported data.
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

5 major / 6 minor

Summary. The paper presents an approach and technical framework for incorporating potentially uncertain human advice into reinforcement learning (RL) for learning complex model transformation (MT) sequences. Advice is expressed through a domain-specific language as values on states, compiled into Subjective Logic (SL) opinions via a base rate, uncertainty calibration, and belief/disbelief computation, then fused with the agent's policy (converted into opinions) using the Belief Constraint Fusion (BCF) operator, and finally transformed back to a probability policy. The framework is implemented on top of EMF and VIATRA. The evaluation compares oracle and single-human advisors at uncertainty levels u = 0.0, 0.2, 0.4, 0.6, 0.8 and various advice quotas on a 12×12 Frozen Lake grid world, reporting cumulative reward and pairwise t-tests. The paper's abstract and conclusion claim that human guidance, even if uncertain, substantially improves RL performance.

Significance. If the central claim held, the paper would be a useful contribution to model-driven engineering and human-in-the-loop RL: domain experts could provide imprecise advice with quantified uncertainty to accelerate the inference of complex MT chains. The paper's strengths include a fully modeled prototype, a publicly available replication package, a systematic sweep of uncertainty levels, and hyperparameter tuning that favors the unadvised baseline. However, the reported evidence supports only low-to-moderate uncertainty (u ≤ 0.4); the headline claim is overstated and internally contradicted by the paper's own statistics. The formalization also contains several underspecified or erroneous steps, notably the neighborhood definition and the advice-to-belief mapping, which need correction. The evaluation on a single toy grid world limits external validity for real MDE problems.

major comments (5)
  1. [VI-B] Sec. VI-B, Table III: the section first states that for u ≥ 0.6 the differences between advised and unadvised agents become statistically insignificant, then concludes that 'human guidance contributes to significant performance improvement in RL-based inference of complex MTs, even at moderate-to-high uncertainty.' These statements are contradictory if 'significant' is used in the statistical sense established by the t-tests. Table III shows at u = 0.8 the Oracle 20% condition (523.6) falls below the unadvised baseline (698.266). The abstract and conclusion repeat the 'even if uncertain' claim. The evidence supports only low-to-moderate uncertainty (u ≤ 0.4), so the headline claim must be qualified in the abstract, conclusion, and Section VI-B.
  2. [V-C-2-a] Sec. V-C-2-a, Definitions 1–2: Definition 1 defines neighbors as states reachable from si via one action, and Definition 2 defines N(si) as the set of such successor states. The subsequent fusion step states that advice about si is fused with opinions of actions 'that lead from its neighboring states N(si) to si,' which requires predecessors of si, not successors. In a directed MDP or general MT setting these sets differ; the definitions as written do not support the fusion step. The notation '∀a∈N(si).A' is also not well-formed. Please correct the definition (or the fusion description) to refer to the set of states from which si is reachable.
  3. [V-B-2] Sec. V-B-2, Eq. (2): the uncertainty calibration is circular. The text reads 'From the identity of ui = 1 − (bi + di), it follows that umax = 1 − (bi + di)', but that identity gives the current uncertainty ui, not an upper bound. Substituting into Eq. (2) yields ui = (δ/δmax)·ui, which is only consistent if δ = δmax. If umax is an independent user-specified constant, the text should say so explicitly. As written, the formula cannot compute ui in the intended way, and the subsequent example bypasses the formula by assigning ui directly.
  4. [V-B-3] Sec. V-B-3, Eqs. (4)–(6): the mapping from the advice value v (e.g., the DSL integer −2..+2) to the index j used in the belief/disbelief formulas is never defined. The example states that v = 2 corresponds to j = 5 for n = 5, but no general mapping is provided. This makes the opinion compilation step underspecified and prevents reproduction in other domains.
  5. [VI] Sec. VI (Evaluation): the empirical evaluation is carried out exclusively on a 12×12 Frozen Lake grid world encoded in EMF, not on complex MT problems such as model synchronization or design-space exploration. The abstract and conclusion generalize to 'complex MTs' and 'efficient development of complex MTs.' While Sec. VI-C acknowledges this as an external validity threat, the central empirical claim is not supported outside the toy domain. A single environment with one reward structure is insufficient evidence for the broad claim made in the title and abstract.
minor comments (6)
  1. [V-B-3, Eq. (4)] Eq. (4) writes bi = (j−1/n−1)×(1−ui), which is ambiguous; the intended fraction is (j−1)/(n−1) as used in Eq. (6).
  2. [Table II] Table II contains a note about additional experiments with cooperating humans; this asserts results that are not presented or statistically analyzed in the paper, so the claim cannot be verified.
  3. [VI-A] The 'single human' advisor condition is implemented by an author with full information of the problem space and synthetic uncertainty values; this is not a realistic uncertain human expert, and the limitation should be stated more prominently.
  4. [VI-B] The paper does not report p-values or effect sizes for the t-tests; the significance statements in Sec. VI-B cannot be checked from the manuscript alone, even though the replication package is cited.
  5. [V-C-2-a] The set notation '∀a∈N(si).A' should be rewritten as a formal quantifier over states and actions, e.g., 'for all s ∈ N(si) and all a ∈ A such that π(a|s) ↦ si'.
  6. [Listing 5] The grammar in Listing 5 uses 'ˆ-?[0-2]', which allows values like −0 and 0, but the mapping to the advice scale is not explained.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central claim is an empirical comparison with advice as an input, and the only self-citation to the authors' prior method is not load-bearing for the reported reward results.

full rationale

The central claim—that human guidance with uncertainty improves RL performance in inferring MT sequences—is established by a controlled comparison in Section VI, where advice is an exogenous input, uncertainty levels are swept synthetically (u = 0.0–0.8), and the RL hyperparameters were deliberately tuned to favor the unadvised baseline. Equations (1)–(6) and the BCF fusion rule define the policy-shaping mechanism; they do not assert the performance result. The outcome is measured empirically against an unadvised agent on the same 12x12 Frozen Lake task, so the conclusion does not reduce to the method's definitions. The one self-citation to [11,32] in Section V imports the general opinion-guided RL method, but the present paper restates the essential equations and, more importantly, evaluates the method against an unadvised baseline within this paper, so the citation does not force the empirical conclusion. The paper does contain an internal consistency problem: Section VI-B reports that for u >= 0.6 the advised-versus-unadvised differences are "statistically insignificant," then concludes that guidance helps "even at moderate-to-high uncertainty." That is an overgeneralization of the evidence and a statistical/correctness concern, not circularity, because the empirical input is not the same as the claimed output. No circular step is exhibited, and the derivation chain is self-contained with respect to circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central empirical comparison does not depend on fitted physical constants; the parameters listed below are experimental settings and hand-chosen scale parameters. None are fitted to force the observed improvement; the uncertainty sweep is a deliberate independent variable.

free parameters (5)
  • Advice value scale length n = 5 (example)
    Equation 6 requires the length of the advice value scale to map the integer advice value v to an index j; n=5 is assumed without a defined procedure.
  • Uncertainty levels u = 0.0, 0.2, 0.4, 0.6, 0.8
    Synthetically set in experiments; not derived from the distance-based calibration in Eq. 2.
  • Advice quotas = 100%, 20% (oracle); 10%, 5% (human)
    Chosen experiment parameters.
  • RL hyperparameters = alpha=0.9, gamma=1.0, episodes=10000, max steps=100
    Chosen and tuned to favor the unadvised baseline, as stated in Sec. VI-A.
  • Environment size and hole ratio = 12x12, 20% holes
    Chosen test configuration, not representative of real MDE tasks.
assumptions (5)
  • standard math Standard RL framework: MDP, policy gradient, and reward accumulation.
    Invoked in Sec. II-A and Sec. VI-A to define the learning setup.
  • standard math Subjective Logic opinion calculus and the Belief Constraint Fusion operator.
    Used in Sec. II-C and Sec. V-C-2-b; formulas are taken from Josang's framework without proof.
  • domain assumption Advice about state benefit can be translated into policy priorities via opinion fusion and normalization.
    This is the central mechanism of Sec. V; no independent evidence links state-level opinions to effective policy shaping in graph-morphism state spaces.
  • domain assumption A reward structure can be formulated for the target MDE problem.
    Stated explicitly in Sec. VI-C-1 as an assumption and limitation.
  • domain assumption Uncertainty can be calibrated by a distance metric with a linear discount function (Eqs. 2-3).
    Proposed in Sec. V-B-2 but never evaluated; experiments use synthetic u values instead.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Complex Model Transformations by Reinforcement Learning with Uncertain Human Guidance." pith.science (2026). https://pith.science/paper/3RR74V76

@misc{pith2026250620883,
  author       = {Pith},
  title        = {Pith review of: Complex Model Transformations by Reinforcement Learning with Uncertain Human Guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3RR74V76}},
  note         = {Machine review of arXiv:2506.20883}
}
read the original abstract

Model-driven engineering problems often require complex model transformations (MTs), i.e., MTs that are chained in extensive sequences. Pertinent examples of such problems include model synchronization, automated model repair, and design space exploration. Manually developing complex MTs is an error-prone and often infeasible process. Reinforcement learning (RL) is an apt way to alleviate these issues. In RL, an autonomous agent explores the state space through trial and error to identify beneficial sequences of actions, such as MTs. However, RL methods exhibit performance issues in complex problems. In these situations, human guidance can be of high utility. In this paper, we present an approach and technical framework for developing complex MT sequences through RL, guided by potentially uncertain human advice. Our framework allows user-defined MTs to be mapped onto RL primitives, and executes them as RL programs to find optimal MT sequences. Our evaluation shows that human guidance, even if uncertain, substantially improves RL performance, and results in more efficient development of complex MTs. Through a trade-off between the certainty and timeliness of human advice, our method takes a step towards RL-driven human-in-the-loop engineering methods.

Figures

Figures reproduced from arXiv: 2506.20883 by the authors.

Figure 1
Figure 1. The Frozen Lake running example at the top leftmost tile (Start). Its aim is to reach the bottom rightmost tile (Goal) while avoiding terminating states (Holes). Model transformations Operational semantics are given by MTs. As shown in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. , moving to the tile to the right is achieved by executing the moveRightRule MT rule, detailed in Listing 1. To illustrate our approach, we use the VIATRA framework [31]. moveRightRule : FrozenTile : Tile t: Truck currentState currentState t: Truck l :Lake currentState agent x = 1 y = 0 t1: FrozenTile x = 2 y = 0 t2: FrozenTile M1 t: Truck l :Lake currentState agent x = 1 y = 0 t1: FrozenTile x = 2 y = 0 t2: FrozenT… view at source ↗
Figure 3
Figure 3. Architectural overview B. Domain-specific (meta)modeling The framework allows for executing MTs as RL programs in an MDE process, with minimal development effort. In step 1③❥, a Domain-specific metamodel is developed by the Domain expert that reuses some key concepts of the RL metamodel. Specifically, this step might be aided by an RL expert in case the domain expert is not familiar with RL concepts. These cases are… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: RL lifecycle for learning MT sequences The execution of the externally registered MTs starts by matching their LHS rules, resulting in some number of Matched LHSs. Typically, multiple MTs’ LHSs are matched, which requires resolving conflicts. In our implementation plat…
Figure 5
Figure 5. Figure 5: Method for incorporating advice into RL grid world of the running example is shown in Listing 5. An example set of advice expressed in the DSL about the running example ( [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Performance comparison of synthetic oracles and real human advisors in terms of cumulative reward [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 47 canonical work pages

  1. [2]

    Recommending model refactoring rules from refactoring examples,

    C. E. Mokaddem, H. Sahraoui, and E. Syriani, “Recommending model refactoring rules from refactoring examples,” in Proc. of the 21th ACM/IEEE Intl. Conference on Model Driven Engineering Languages and Systems , ser. MODELS ’18, Copenhagen, Denmark: ACM, 2018, pp. 257–266. DOI: 10.1145/3239372.3239406

    work page arXiv 2018

  2. [3]

    Multi-objective optimization in rule-based design space exploration,

    H. Abdeen, D. Varr ´o, H. Sahraoui, A. S. Nagy, C. Debreceni, ´A. Heged¨us, and ´A. Horv ´ath, “Multi-objective optimization in rule-based design space exploration,” in Proc. of the 29th ACM/IEEE Intl Confer- ence on Automated Software Engineering , ACM, 2014, pp. 289–300. DOI: 10.1145/2642937.2643005

    work page arXiv 2014

  3. [4]

    R. S. Sutton and A. G. Barto, Reinforcement learning: An introduction . MIT press, 2018

    2018

  4. [5]

    Personalized and automatic model repairing using reinforcement learning,

    A. Barriga, A. Rutle, and R. Heldal, “Personalized and automatic model repairing using reinforcement learning,” in 2019 ACM/IEEE 22nd International Conference on Model Driven Engineering Languages and Systems Companion (MODELS-C) , 2019, pp. 175–181. DOI: 10.1109/ MODELS-C.2019.00030

    arXiv 2019

  5. [6]

    Towards reinforcement learning for in-place model transformations,

    M. Eisenberg, H.-P. Pichler, A. Garmendia, and M. Wimmer, “Towards reinforcement learning for in-place model transformations,” in 2021 ACM/IEEE 24th Intl Conf. on Model Driven Engineering Languages and Systems (MODELS) , 2021, pp. 82–88

    work page 2021

  6. [7]

    Facil- itating the migration to the microservice architecture via model-driven reverse engineering and reinforcement learning,

    M. Dehghani, S. Kolahdouz-Rahimi, M. Tisi, and D. Tamzalit, “Facil- itating the migration to the microservice architecture via model-driven reverse engineering and reinforcement learning,”Soft. Sys. Mod., vol. 21, no. 3, pp. 1115–1133, 2022. DOI: 10.1007/s10270-022-00977-3

    work page doi:10.1007/s10270-022-00977-3 2022

  7. [8]

    Reinforcement learning with human advice: A survey,

    A. Najar and M. Chetouani, “Reinforcement learning with human advice: A survey,” Frontiers in Robotics and AI , vol. 8, 2021, ISSN : 2296-9144. DOI: 10.3389/frobt.2021.584075

    work page arXiv 2021

  8. [9]

    Parmorel: A framework for customizable model repair,

    A. Barriga, R. Heldal, A. Rutle, and L. Iovino, “Parmorel: A framework for customizable model repair,” Soft. Sys. Mod., vol. 21, no. 5, pp. 1739– 1762, 2022

    work page 2022

Show all 58 references
  1. [10]

    A model-driven framework for guided design space exploration,

    ´A. Heged ¨us, ´A. Horv ´ath, and D. Varr ´o, “A model-driven framework for guided design space exploration,” Autom Softw Eng , vol. 22, no. 3, pp. 399–436, 2015. DOI: 10.1007/s10515-014-0163-1

    2015 doi

  2. [11]

    Towards model repair by human opinion–guided rein- forcement learning,

    K. Dagenais, “Towards model repair by human opinion–guided rein- forcement learning,” in Proceedings of the ACM/IEEE 27th International Conference on Model Driven Engineering Languages and Systems , ser. MODELS Companion ’24, Linz, Austria: ACM, 2024, pp. 192–195. DOI: 10.1145...

    2024

  3. [12]

    Markov decision processes,

    M. L. Puterman, “Markov decision processes,” Handbooks in operations research and management science , vol. 2, pp. 331–434, 1990

    1990

  4. [13]

    Exploring fault parameter space using reinforcement learning- based fault injection,

    M. Moradi, B. J. Oakes, M. Saraoglu, A. Morozov, K. Janschek, and J. Denil, “Exploring fault parameter space using reinforcement learning- based fault injection,” in 2020 50th Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN-W) , 2020...

    2020

  5. [14]

    DEVS model construction as a reinforcement learning problem,

    I. David and E. Syriani, “DEVS model construction as a reinforcement learning problem,” in 2022 Annual Modeling and Simulation Conference (ANNSIM), 2022, pp. 30–41. DOI: 10 . 23919 / ANNSIM55834 . 2022 . 9859369

    2022

  6. [15]

    Agent- advising approaches in an interactive reinforcement learning scenario,

    F. Cruz, P. W ¨uppen, S. Magg, A. Fazrie, and S. Wermter, “Agent- advising approaches in an interactive reinforcement learning scenario,” in 2017 Joint IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL-EpiRob) , 2017, pp. 209–214. DOI: 10....

    2017

  7. [16]

    Training a robot with evaluative feedback and unlabeled guidance signals,

    A. Najar, O. Sigaud, and M. Chetouani, “Training a robot with evaluative feedback and unlabeled guidance signals,” in 2016 25th IEEE Inter- national Symposium on Robot and Human Interactive Communication (RO-MAN), 2016, pp. 261–266. DOI: 10.1109/ROMAN.2016.7745140

    2016

  8. [17]

    Dynamic reward shaping: Training a robot by voice,

    A. C. Tenorio-Gonzalez, E. F. Morales, and L. Villase ˜nor-Pineda, “Dynamic reward shaping: Training a robot by voice,” in Advances in Artificial Intelligence – IBERAMIA 2010 , Springer, 2010, pp. 483–492, ISBN : 978-3-642-16952-6

    2010

  9. [18]

    Reinforcement learning with human teachers: Evidence of feedback and guidance with implications for learning performance,

    A. L. Thomaz and C. Breazeal, “Reinforcement learning with human teachers: Evidence of feedback and guidance with implications for learning performance,” in Proc of the 21st National Conf on Artificial Intelligence - Volume 1 , ser. AAAI’06, Boston, Massachusetts: AAAI Press, ...

    2006

  10. [19]

    Policy shaping: Integrating human feedback with reinforcement learn- ing,

    S. Griffith, K. Subramanian, J. Scholz, C. L. Isbell, and A. L. Thomaz, “Policy shaping: Integrating human feedback with reinforcement learn- ing,” in Advances in Neural Information Processing Systems , vol. 26, 2013

    2013

  11. [20]

    Jøsang, Subjective logic

    A. Jøsang, Subjective logic. Springer, 2016, vol. 3

    2016

  12. [21]

    Expressing confidence in models and in model transformation elements,

    L. Burgue ˜no, M. F. Bertoa, N. Moreno, and A. Vallecillo, “Expressing confidence in models and in model transformation elements,” in Proc of the 21th ACM/IEEE Intl Conference on Model Driven Engineering Languages and Systems , ACM, 2018, pp. 57–66

    2018

  13. [22]

    A belief-theoretic framework for the collaborative development and integration of para-consistent conceptual models,

    E. Bagheri and A. A. Ghorbani, “A belief-theoretic framework for the collaborative development and integration of para-consistent conceptual models,” J. Sys. & Soft. , vol. 82, no. 4, pp. 707–729, 2009, ISSN : 0164-

    2009

  14. [23]

    Uncertainty representation in software models: A survey,

    J. Troya, N. Moreno, M. F. Bertoa, and A. Vallecillo, “Uncertainty representation in software models: A survey,” Soft. Sys. Mod. , vol. 20, no. 4, pp. 1183–1213, 2021, ISSN : 1619-1374. DOI: 10.1007/s10270- 020-00842-1

    2021 doi

  15. [24]

    Uncertainty-aware consistency checking in industrial settings,

    R. Jongeling and A. Vallecillo, “Uncertainty-aware consistency checking in industrial settings,” in ACM/IEEE International Conference on Model Driven Engineering Languages and Systems , 2023, pp. 73–83

    2023

  16. [25]

    Model repair with quality-based reinforcement learning.,

    L. Iovino, B. Angela, R. Adrian, and H. Rogardt, “Model repair with quality-based reinforcement learning.,” JOT, vol. 19, no. 2, 17:1, 2020. DOI: 10.5381/jot.2020.19.2.a17

    2020 doi

  17. [26]

    From single-objective to multi-objective reinforcement learning-based model transformation,

    M. Eisenberg and M. Wimmer, “From single-objective to multi-objective reinforcement learning-based model transformation,” Soft. Sys. Mod. , 2024, ISSN : 1619-1374. DOI: 10.1007/s10270-024-01233-6

    2024 doi

  18. [27]

    A tool for automatically selecting optimal model transformation chains,

    F. Basciani, D. Di Ruscio, M. D’Emidio, D. Frigioni, A. Pierantonio, and L. Iovino, “A tool for automatically selecting optimal model transformation chains,” in Proceedings of the 21st ACM/IEEE Intl. Conf. on Model Driven Engineering Languages and Systems: Companion Proceeding...

    2018

  19. [28]

    Multi-objective model transformation chain exploration with MOMoT,

    M. Eisenberg, A. Sahay, D. Di Ruscio, L. Iovino, M. Wimmer, and A. Pierantonio, “Multi-objective model transformation chain exploration with MOMoT,” Inf Softw Technol , vol. 174, p. 107 500, 2024, ISSN : 0950-5849

    2024

  20. [29]

    Graph transformation in a nutshell,

    R. Heckel, “Graph transformation in a nutshell,” Electronic Notes in Theoretical Computer Science, vol. 148, no. 1, pp. 187–198, 2006, ISSN : 1571-0661. DOI: https://doi.org/10.1016/j.entcs.2005.12.018

    2006 doi

  21. [30]

    Domain-specific model distance measures,

    E. Syriani, R. Bill, and M. Wimmer, “Domain-specific model distance measures,” Journal of Object Technology , vol. 18, no. 3, 3:1–19, 2019, The 12th International Conference on Model Transformations, ISSN : 1660-1769. DOI: 10.5381/jot.2019.18.3.a3

    2019 doi

  22. [31]

    VIATRA 3: A reactive model transformation platform,

    G. Bergmann, I. David, ´A. Heged ¨us, ´A. Horv ´ath, I. R ´ath, Z. Ujhelyi, and D. Varr ´o, “VIATRA 3: A reactive model transformation platform,” in Theory and Practice of Model Transformations - 8th International Conference, ICMTSTAF 2015, L’Aquila, Italy. Proceedings , ser. ...

    2015

  23. [32]

    Opinion-guided reinforcement learning,

    K. Dagenais and I. David, “Opinion-guided reinforcement learning,” Tech. Rep., 2024. arXiv: 2405.17287 [cs.LG]

    2024 arXiv

  24. [33]

    Model-driven engineering,

    D. C. Schmidt, “Model-driven engineering,” Computer-IEEE Computer Society, vol. 39, no. 2, p. 25, 2006

    2006

  25. [34]

    Jøsang, Subjective Logic

    A. Jøsang, Subjective Logic . Springer International Publishing, 2016, ISBN : 9783319423371. DOI: 10.1007/978-3-319-42337-1

    2016 doi

  26. [35]

    Automatic resolution of model merging conflicts using quality-based reinforcement learning,

    M. Sharbaf, B. Zamani, and G. Suny ´e, “Automatic resolution of model merging conflicts using quality-based reinforcement learning,” Journal of Computer Languages, vol. 71, p. 101 123, 2022. DOI: https://doi.org/ 10.1016/j.cola.2022.101123

    2022

  27. [36]

    Incremental resolution of model inconsistencies,

    T. Mens and R. Van Der Straeten, “Incremental resolution of model inconsistencies,” in Recent Trends in Algebraic Development Techniques, Springer, 2007, pp. 111–126, ISBN : 978-3-540-71998-4

    2007

  28. [37]

    A mde approach for heterogeneous models consistency,

    M. El Hamlaoui, S. Bennani, M. Nassar, S. Ebersold, and B. Coulette, “A mde approach for heterogeneous models consistency,” in Proc of the 13th Intl Conference on Evaluation of Novel Approaches to Software Engineering , ser. ENASE 2018, Funchal, Madeira, Portugal: SCITEPRESS, ...

    2018 doi

  29. [38]

    Engineering process transformation to manage (in)consistency,

    I. David, J. Denil, K. Gadeyne, and H. Vangheluwe, “Engineering process transformation to manage (in)consistency,” in Proceedings of the 1st International Workshop on Collaborative Modelling in MDE (COMMitMDE 2016) co-located with ACM/IEEE 19th International Conference on Mode...

    2016

  30. [40]

    Model consistency as a heuristic for eventual correctness,

    I. David, H. Vangheluwe, and E. Syriani, “Model consistency as a heuristic for eventual correctness,” Journal of Computer Languages , vol. 76, p. 101 223, 2023, ISSN : 2590-1184. DOI: https://doi.org/10. 1016/j.cola.2023.101223

    2023

  31. [41]

    Driving requirements evolution by engineers’ opinions,

    K. Dagenais and I. David, “Driving requirements evolution by engineers’ opinions,” in ACM/IEEE International Conference on Model Driven Engineering Languages and Systems Companion, MODELS-C , ACM,

  32. [42]

    Change-preserving model repair,

    G. Taentzer, M. Ohrndorf, Y . Lamo, and A. Rutle, “Change-preserving model repair,” in Fundamental Approaches to Software Engineering , Springer, 2017, pp. 283–299, ISBN : 978-3-662-54494-5

    2017

  33. [43]

    Model-transformation-based computational design synthesis for mis- sion architecture optimization,

    S. J. I. Herzig, S. Mandutianu, H. Kim, S. Hernandez, and T. Imken, “Model-transformation-based computational design synthesis for mis- sion architecture optimization,” in 2017 IEEE Aerospace Conference , 2017, pp. 1–15. DOI: 10.1109/AERO.2017.7943953

    2017

  34. [44]

    Probabilistic graph queries for design space exploration under uncertainty,

    M. F ¨oldi´ak, “Probabilistic graph queries for design space exploration under uncertainty,” in Proceedings of the ACM/IEEE 27th International Conference on Model Driven Engineering Languages and Systems , ser. MODELS Companion ’24, Linz, Austria: ACM, 2024, pp. 142–148, ISBN ...

    2024

  35. [45]

    Grand challenges in model-driven engineering: An analysis of the state of the research,

    A. Bucchiarone, J. Cabot, R. F. Paige, and A. Pierantonio, “Grand challenges in model-driven engineering: An analysis of the state of the research,” Soft. Sys. Mod. , vol. 19, no. 1, pp. 5–13, 2020, ISSN : 1619-

    2020

  36. [46]

    Marlon: A domain-specific language for multi-agent reinforcement learning on networks,

    T. Molderez, B. Oeyen, C. De Roover, and W. De Meuter, “Marlon: A domain-specific language for multi-agent reinforcement learning on networks,” in Proc of the 34th ACM/SIGAPP Symposium on Applied Computing, ACM, 2019, pp. 1322–1329. DOI: 10 . 1145 / 3297280 . 3297413

    2019

  37. [47]

    Ontology-based development of domain-specific languages via customizing base language,

    G. Kulagin, I. Ermakov, and L. Lyadova, “Ontology-based development of domain-specific languages via customizing base language,” in 2022 IEEE 16th International Conference on Application of Information and Communication Technologies (AICT) , 2022, pp. 1–6. DOI: 10 . 1109 / AIC...

    2022

  38. [48]

    Automated generation of testing tools for domain- specific languages,

    H. Wu and J. Gray, “Automated generation of testing tools for domain- specific languages,” in Proceedings of the 20th IEEE/ACM International Conference on Automated Software Engineering , ACM, 2005, pp. 436–

    2005

  39. [49]

    Generation of custom textual model editors,

    E. Syriani, D. Riegelhaupt, B. Barroca, and I. David, “Generation of custom textual model editors,” Modelling, vol. 2, no. 4, pp. 609–625, 2021, ISSN : 2673-3951. DOI: 10.3390/modelling2040032

    2021 doi

  40. [50]

    Toward intelligent generation of tailored graphical concrete syntax,

    M. Ben Chaaben, O. Ben Sghaier, M. Dhaouadi, N. Elrasheed, I. Darif, I. Jaoua, B. Oakes, E. Syriani, and M. Hamdaqa, “Toward intelligent generation of tailored graphical concrete syntax,” in Proceedings of the ACM/IEEE 27th International Conference on Model Driven Engineering ...

    2024

  41. [51]

    Adaptive modelling languages: Abstract syntax and model migration,

    J. de Lara and E. Guerra, “Adaptive modelling languages: Abstract syntax and model migration,” ACM Trans. Softw. Eng. Methodol. , vol. 34, no. 3, 2025, ISSN : 1049-331X. DOI: 10.1145/3702975

    2025 doi

  42. [52]

    Shajari and I

    B. Shajari and I. David, Bridging the silos of digitalization and sustain- ability by twin transition: A multivocal literature review , 2025. arXiv: 2506.04267 [physics.soc-ph]. [Online]. Available: https://arxiv. org/abs/2506.04267

    2025 arXiv

  43. [53]

    User experience for model-driven engineering: Chal- lenges and future directions,

    S. Abrah ˜ao, F. Bourdeleau, B. Cheng, S. Kokaly, R. Paige, H. St ¨oerrle, and J. Whittle, “User experience for model-driven engineering: Chal- lenges and future directions,” in 2017 ACM/IEEE 20th International Conference on Model Driven Engineering Languages and Systems (MODE...

    2017 doi

  44. [54]

    Blended modeling in commercial and open-source model-driven software engineering tools: A systematic study,

    I. David, M. Latifaj, J. Pietron, W. Zhang, F. Ciccozzi, I. Malavolta, A. Raschke, J.-P. Stegh ¨ofer, and R. Hebig, “Blended modeling in commercial and open-source model-driven software engineering tools: A systematic study,” Soft Sys Mod , vol. 22, no. 1, pp. 415–447, 2023. D...

    2023 doi

  45. [55]

    Statecharts: A visual formalism for complex systems,

    D. Harel, “Statecharts: A visual formalism for complex systems,” Science of Computer Programming , vol. 8, no. 3, pp. 231–274, 1987. DOI: https://doi.org/10.1016/0167-6423(87)90035-9

    1987 doi

  46. [56]

    Automatic code generation from unified modelling language sequence diagrams,

    D. Kundu, D. Samanta, and R. Mall, “Automatic code generation from unified modelling language sequence diagrams,” IET Software, vol. 7, no. 1, pp. 12–28, 2013. DOI: https://doi.org/10.1049/iet-sen.2011.0080

    2013

  47. [439]

    DOI: 10.1145/1101908.1101993

    arXiv

  48. [1212]

    DOI: https://doi.org/10.1016/j.jss.2008.10.012

    2008 doi

  49. [1374]

    DOI: 10.1007/s10270-019-00773-6

    doi

  50. [2024]

    DOI: 10.1145/3652620.3688566

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.