Pith. sign in

REVIEW 2 major objections 5 minor 39 references

A graph neural net plus verifiers synthesizes enforceable job-level dependencies for automotive cause-effect chains faster and more often than the classic greedy heuristic.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 08:27 UTC pith:M5VHPVFF

load-bearing objection Solid first ML attack on JLD synthesis: Generate-and-Verify + DP checker + GNN beats the classic greedy on success and runtime, with the usual synthetic-data caveats. the 2 major comments →

arxiv 2607.02624 v1 pith:M5VHPVFF submitted 2026-07-02 cs.SE cs.LGcs.SYeess.SY

Schedulable Job-Level Dependencies for Cause-Effect Chains via Graph Neural Networks

Silviu S. Craciunas , Christian Hakert , Jian-Jia Chen , Zden\v{e}k Hanz\'alek , Paul Pop This is my paper
classification cs.SE cs.LGcs.SYeess.SY
keywords cause-effect chainsjob-level dependenciesgraph neural networksEDF schedulingdata agegenerate-and-verifyreal-time systemsautomotive software
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Automotive software must keep sensor-to-actuator data fresh under strict end-to-end age budgets. Job-level dependencies (JLDs) prune the worst-case paths without fixing a concrete schedule, but earlier synthesis methods only grow the dependency set greedily and never check whether a real scheduler can enforce it or whether the whole task set remains schedulable. This paper trains a two-level graph neural network to propose JLD candidates from the structural patterns of periods and chain topology, then wraps the network in a Generate-and-Verify loop: a dynamic-programming age checker, a per-chain EDF feasibility transform, and a system-level demand-bound test accept or reject every proposal. On 1 500 synthetic use-cases the learned generator raises the fraction of fully valid systems well above the original greedy baseline while cutting median per-chain synthesis time by orders of magnitude, showing that learned structural priors can replace exponential propagation-tree search for this class of real-time problems.

Core claim

A two-level GNN with temperature-controlled sampling, embedded in a Generate-and-Verify architecture that uses a safe DP data-age checker plus EDF feasibility tests, produces JLD sets that are simultaneously age-correct, EDF-enforceable and system-schedulable more often than the classic greedy heuristic, and does so at far lower synthesis cost.

What carries the argument

The Generate-and-Verify loop: a temperature-sampled two-level GNN proposes candidate JLDs; a polynomial DP age checker, a per-chain EDF precedence transform and a system-level demand-bound test accept or reject each candidate until a jointly feasible set is found (or a best-effort set is retained).

Load-bearing premise

The synthetic labels produced by the greedy heuristic and a handful of randomized seeds already contain the structural patterns needed for the network to generalize to unseen period sets and chain shapes.

What would settle it

Train the same GNN architecture on the same volume of data drawn only from the original greedy heuristic (no randomized seeds) and measure whether fully-valid success on the Unknown period set collapses relative to the paper's reported figures; a large drop would falsify the claim that the learned priors are robust.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper addresses synthesis of job-level dependencies (JLDs) that bound worst-case data age of multi-rate cause-effect chains while remaining enforceable under EDF and jointly schedulable. It introduces a Generate-and-Verify loop that pairs candidate generators with a polynomial DP data-age checker (Algorithm 2, Lemmas 1–2, Theorem 3), a per-chain EDF precedence transform (four cases C.1–C.4), and a system-level demand-bound test. Two generators are proposed: GreedyRand-JLD (randomized multi-JLD-per-pair extension of Becker et al.) and GNN-JLD (two-level GNN for edge selection + MLP for position selection with temperature-controlled sampling). On 1 500 synthetic use-cases spanning Known/Similar/Unknown period regimes, GNN-JLD (K=50) raises fully-valid use-cases to 97.4/86.8/83.2 % versus 76.8/56.4/59.6 % for single-shot Greedy-JLD, at median per-chain generation times of 6–8 ms (orders-of-magnitude faster in the tail).

Significance. The work closes a concrete gap left by Becker et al.: prior JLD synthesis ignored enforceability and system-level schedulability. The Generate-and-Verify architecture, the DP checker with accompanying correctness argument, and the multi-regime evaluation (including K=1 ablations and a pure-random control) constitute a solid engineering contribution. The demonstration that a learned structural prior can replace exponential propagation-tree enumeration for this class of real-time problems is of practical interest for automotive tool chains and of methodological interest for learning-assisted real-time design. Strengths include the explicit verifier guarantees, the Unknown-set generalization experiment, and the transparent accounting of best-effort candidates.

major comments (2)
  1. §5.2 training pipeline: labels are produced almost exclusively by the same greedy family (77.1 % Becker, 15.1 % GreedyRand) that the paper later claims to outperform. While the Unknown-set results and K=1 ablations mitigate circularity concerns, the manuscript should quantify how much of the reported gain is attributable to the richer multi-JLD-per-pair encoding versus genuine structural learning. A short experiment that trains Level 1/2 exclusively on pure-random or on optimally enumerated short-chain labels (or an explicit statement that such data are unavailable) would strengthen the central claim that “learned structural priors” replace enumeration.
  2. §6.2 / Table 2 and Figure 8: GNN-JLD produces substantially more JLDs per chain (mean 10–15, long tail >50) than Greedy-JLD (3.6–5.1). The paper correctly notes that some of these configurations fail the EDF transform. The evaluation reports only success rates; it does not report the distribution of residual normalized data age or the fraction of over-constrained (zero-path) chains among the rejected candidates. Adding these secondary metrics would clarify whether the higher JLD count is a benign side-effect of chain-HP indexing or a systematic over-pruning risk that practitioners must manage.
minor comments (5)
  1. §2.1 / footnote 2: the claim that the method is independent of publication semantics is plausible but never demonstrated; a one-sentence remark on how Dmin would be redefined under LET or BCET would help readers.
  2. Algorithm 2 line 11: the cascade formula d′_next ← max(d′, Rmin) + C differs slightly from the prose Eq. (1); a short clarification that the two are equivalent under the delayed-publication model would avoid confusion.
  3. Figure 1 caption and §5: the temperature ladder is listed but the precise mapping from attempt index k to θ is only described in prose; a small table or inline formula would improve reproducibility.
  4. Table 1: the O(m N²) entry for GNN generation absorbs the GNN forward-pass cost; stating the hidden-dimension dependence (or noting that it is dominated) would make the complexity claim fully transparent.
  5. §7: the related-work discussion of ML-for-scheduling is adequate but omits recent GNN-based job-shop surveys that would situate the two-level architecture more clearly.

Circularity Check

0 steps flagged

No significant circularity: empirical GNN generator trained on heuristic labels, verified by independent classical DP/EDF checkers, and evaluated on held-out/unseen benchmarks.

full rationale

The paper's central claim is an empirical performance result (higher fully-valid use-case rates and orders-of-magnitude lower synthesis time than Greedy-JLD) obtained inside a Generate-and-Verify loop. Training labels are produced by the same family of greedy heuristics (Becker et al. plus GreedyRand fallbacks), yet this does not create a definitional or fitted-input circularity: every candidate is accepted or rejected by a separately proved DP data-age checker (Lemmas 1-2, Theorem 3) and classical EDF demand-bound / precedence-transform tests that do not depend on the GNN or on the training labels. Evaluation metrics are measured on Known/Similar/Unknown partitions (including completely unseen period sets) and are further isolated by K=1 argmax and pure-random ablations. No equation equates a 'prediction' to a fitted constant, no uniqueness theorem is imported from the authors, and no self-citation is load-bearing for the correctness or the reported gains. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 3 invented entities

The central empirical claim rests on standard real-time scheduling axioms (EDF demand-bound test, job-level precedence transformation), the data-propagation model of Becker et al., and a collection of free architectural and training hyper-parameters. No new physical entities are postulated; the GNN and Generate-and-Verify loop are algorithmic inventions whose independent evidence is the reported benchmark performance.

free parameters (5)
  • temperature ladder θ = {0.01,0.1,0.3,0.5,0.8,1.0,1.5,2.0,3.0,5.0}
    Ten fixed temperatures {0.01 … 5.0} plus upward shift on system retries; chosen by hand to balance exploitation/exploration.
  • K attempts / S system retries = K=10/50, S=3
    Search budget (K=10 or 50, S=3) that directly affects reported success rates.
  • GNN/MLP hidden dimensions and layer counts
    64-dim node embeddings, 4 message-passing layers, edge head 137 o64 o32 o2, position MLP 7 o32 o32 o1; architectural choices that influence capacity.
  • training epochs and loss weights = 1000 / 2000 / 0.1
    1000 epochs Level-1 (CE + 0.1 MSE), 2000 epochs Level-2 (BCE); weight 0.1 on auxiliary age regression is free.
  • max_iter for GreedyRand = 50000
    Hard limit of 50 000 iterations before giving up on a chain.
axioms (5)
  • domain assumption EDF demand-bound test of Baruah et al. / Pellizzoni & Lipari correctly decides schedulability of the transformed virtual-task set.
    Invoked as the final system-level filter (§3, Eq. 2).
  • domain assumption Data-propagation intervals under delayed publication (or fixed execution time) correctly capture worst-case data age (Becker et al. model).
    Foundation of both the DP checker and the training labels (§2.1).
  • domain assumption Implicit deadlines Di = Ti; constrained-deadline support is claimed but not evaluated.
    Stated for fair comparison with Becker; deadline-ratio feature is present but unused (§2.1 footnote).
  • domain assumption Four cases of offset/deadline adjustment correctly enforce a JLD under preemptive EDF without introducing new cycles.
    Transformation rules C.1–C.4 and topological sort (§3).
  • standard math Dominance pruning of DP states preserves the longest valid path (Lemma 1–2, Theorem 3).
    Proved by induction in §4.1.
invented entities (3)
  • Two-level GNN + temperature-controlled sampler for JLD synthesis independent evidence
    purpose: Generate candidate JLD sets that capture structural period/chain patterns without exponential tree enumeration.
    Core algorithmic contribution; independent evidence is the reported outperformance on held-out and unknown period sets.
  • Generate-and-Verify architecture for JLDs independent evidence
    purpose: Guarantee that every accepted candidate satisfies data-age, per-chain EDF enforceability and system-level schedulability.
    Wrapper that turns an approximate learned proposer into a sound synthesizer; evidence is the multi-level success metrics.
  • DP data-age checker (Algorithm 2) independent evidence
    purpose: Polynomial-time verification of worst-case data age under a candidate JLD set.
    New algorithm with accompanying correctness proof; replaces exponential enumeration for the verification side.

pith-pipeline@v1.1.0-grok45 · 31325 in / 3325 out tokens · 31455 ms · 2026-07-12T08:27:05.423339+00:00 · methodology

0 comments
read the original abstract

Modern automotive software architectures comprise large sets of mixed-criticality functions executing on shared multi-core platforms with strict real-time and end-to-end timing requirements. Sensor-to-actuator data propagation in such systems is typically expressed via cause-effect chains with worst-case data-age budgets. Job-level dependencies (JLDs) have been introduced to provide a schedule-agnostic mechanism for bounding the data age independently of the underlying scheduler. The state-of-the-art methods for synthesizing JLDs, however, do not check whether the produced JLDs are enforceable under a concrete scheduling policy or jointly schedulable at the system level. In this paper we propose the first machine-learning-based JLD synthesis method, built around a two-level Graph Neural Network with temperature-controlled sampling that learns the structural patterns connecting cause-effect chain configurations to their JLD solutions. Since learned outputs may not be correct by construction, we embed the GNN in a novel Generate-and-Verify architecture in which a safe DP data-age checker, together with a per-chain EDF feasibility checker and a system-level demand-bound test, accept or reject each candidate. We show that the ML-based generator substantially outperforms the original greedy heuristic while achieving orders-of-magnitude lower synthesis time, demonstrating that learned structural priors can effectively replace exponential propagation-tree enumeration on this class of real-time scheduling problems.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 27 canonical work pages

  1. [1]

    Learning-assisted schedulability analysis: opportunities and limitations

    Sanjoy Baruah, Pontus Ekberg, and Marion Sudvarg. Learning-assisted schedulability analysis: opportunities and limitations. Real-Time Syst. , 61(3–4):332–358, July 2025. https://doi.org/10.1007/s11241-025-09450-y doi:10.1007/s11241-025-09450-y

    work page doi:10.1007/s11241-025-09450-y 2025

  2. [2]

    Baruah, Louis E

    Sanjoy K. Baruah, Louis E. Rosier, and R. R. Howell. Algorithms and complexity concerning the preemptive scheduling of periodic, real-time tasks on one processor. Real-Time Syst. , 2(4), 1990. https://doi.org/10.1007/BF01995675 doi:10.1007/BF01995675

    work page doi:10.1007/bf01995675 1990

  3. [3]

    Meeting job-level dependencies by task merging

    Matthias Becker. Meeting job-level dependencies by task merging. In 2024 29th Asia and South Pacific Design Automation Conference (ASP-DAC) , pages 792--798, 2024. https://doi.org/10.1109/ASP-DAC58780.2024.10473901 doi:10.1109/ASP-DAC58780.2024.10473901

    doi:10.1109/asp-dac58780.2024.10473901 2024

  4. [4]

    Mechaniser-a timing analysis and synthesis tool for multi-rate effect chains with job-level dependencies

    Matthias Becker, Dakshina Dasari, Saad Mubeen, Moris Behnam, and Thomas Nolte. Mechaniser-a timing analysis and synthesis tool for multi-rate effect chains with job-level dependencies. In Proc. WATERS , 2016

    2016

  5. [5]

    Synthesizing job-level dependencies for automotive multi-rate effect chains

    Matthias Becker, Dakshina Dasari, Saad Mubeen, Moris Behnam, and Thomas Nolte. Synthesizing job-level dependencies for automotive multi-rate effect chains. In Proc. RTCSA , 2016. https://doi.org/10.1109/RTCSA.2016.41 doi:10.1109/RTCSA.2016.41

    work page doi:10.1109/rtcsa.2016.41 2016

  6. [6]

    End-to-end timing analysis of cause-effect chains in automotive embedded systems

    Matthias Becker, Dakshina Dasari, Saad Mubeen, Moris Behnam, and Thomas Nolte. End-to-end timing analysis of cause-effect chains in automotive embedded systems. J. Syst. Archit. , 80, 2017. https://doi.org/10.1016/j.sysarc.2017.09.004 doi:10.1016/j.sysarc.2017.09.004

    work page doi:10.1016/j.sysarc.2017.09.004 2017

  7. [7]

    Analyzing end-to-end delays in automotive systems at various levels of timing information

    Matthias Becker, Dakshina Dasari, Saad Mubeen, Moris Behnam, and Thomas Nolte. Analyzing end-to-end delays in automotive systems at various levels of timing information. SIGBED Rev. , 14(4):8–13, January 2018. https://doi.org/10.1145/3177803.3177805 doi:10.1145/3177803.3177805

    work page doi:10.1145/3177803.3177805 2018

  8. [8]

    Buttazzo

    Enrico Bini and Giorgio C. Buttazzo. Measuring the performance of schedulability tests. Real-Time Syst. , 30(1–2):129–154, May 2005. https://doi.org/10.1007/s11241-005-0507-9 doi:10.1007/s11241-005-0507-9

    work page doi:10.1007/s11241-005-0507-9 2005

  9. [9]

    Deep learning-driven scheduling algorithm for a single machine problem minimizing the total tardiness

    Michal Bouška, Přemysl Šůcha, Antonín Novák, and Zdeněk Hanzálek. Deep learning-driven scheduling algorithm for a single machine problem minimizing the total tardiness. European Journal of Operational Research , 308(3):990--1006, 2023. https://doi.org/10.1016/j.ejor.2022.11.034 doi:10.1016/j.ejor.2022.11.034

    doi:10.1016/j.ejor.2022.11.034 2023

  10. [10]

    Brandenburg and Mahircan Gül

    Björn B. Brandenburg and Mahircan Gül. Global scheduling not required: Simple, near-optimal multiprocessor real-time scheduling with semi-partitioned reservations. In Proc. RTSS , pages 99--110, 2016. https://doi.org/10.1109/RTSS.2016.019 doi:10.1109/RTSS.2016.019

    work page doi:10.1109/rtss.2016.019 2016

  11. [11]

    To MILP or not to MILP ? on AI techniques for the design and optimization of real-time systems

    Daniel Casini. To MILP or not to MILP ? on AI techniques for the design and optimization of real-time systems. Real-Time Syst. , 61(2):294–299, May 2025. https://doi.org/10.1007/s11241-025-09440-0 doi:10.1007/s11241-025-09440-0

    work page doi:10.1007/s11241-025-09440-0 2025

  12. [12]

    Chetto, M

    H. Chetto, M. Silly, and T. Bouchentouf. Dynamic scheduling of real-time tasks under precedence constraints. Real-Time Syst. , 2, 1990. https://doi.org/10.1007/BF00365326 doi:10.1007/BF00365326

    work page doi:10.1007/bf00365326 1990

  13. [13]

    Craciunas, Ramon Serna Oliver , and Valentin Ecker

    Silviu S. Craciunas, Ramon Serna Oliver , and Valentin Ecker. Optimal static scheduling of real-time tasks on distributed time-triggered networked systems. In Proc. ETFA , pages 1--8, 2014. https://doi.org/10.1109/ETFA.2014.7005128 doi:10.1109/ETFA.2014.7005128

    work page doi:10.1109/etfa.2014.7005128 2014

  14. [14]

    u rr, Georg Von Der Br\

    Marco D\" u rr, Georg Von Der Br\" u ggen, Kuan-Hsun Chen, and Jian-Jia Chen. End-to-end timing analysis of sporadic cause-effect chains in distributed systems. ACM Trans. Embed. Comput. Syst. , 18(5s), October 2019. https://doi.org/10.1145/3358181 doi:10.1145/3358181

    work page doi:10.1145/3358181 2019

  15. [15]

    Craciunas, and Marc Boyer

    Anaïs Finzi, Silviu S. Craciunas, and Marc Boyer. Integrating sporadic events in time-triggered systems via affine envelope approximations. In Proc. RTAS , 2024. https://doi.org/10.1109/RTAS61025.2024.00010 doi:10.1109/RTAS61025.2024.00010

    doi:10.1109/rtas61025.2024.00010 2024

  16. [16]

    Scheduling dependent periodic tasks without synchronization mechanisms

    Julien Forget, Frédéric Boniol, Emmanuel Grolleau, David Lesens, and Claire Pagetti. Scheduling dependent periodic tasks without synchronization mechanisms. In Proc. RTAS , pages 301--310, 2010. https://doi.org/10.1109/RTAS.2010.26 doi:10.1109/RTAS.2010.26

    work page doi:10.1109/rtas.2010.26 2010

  17. [17]

    Periodic chains scheduling on dedicated resources - a crucial problem in time-sensitive networks

    Josef Grus, Claire Hanen, and Zden e k Hanz\' a lek. Periodic chains scheduling on dedicated resources - a crucial problem in time-sensitive networks. Comput. Oper. Res. , 180(C), August 2025. https://doi.org/10.1016/j.cor.2025.107072 doi:10.1016/j.cor.2025.107072

    doi:10.1016/j.cor.2025.107072 2025

  18. [18]

    Packing-inspired algorithms for periodic scheduling problems with harmonic periods

    Josef Grus, Claire Hanen, and Zdeněk Hanzálek. Packing-inspired algorithms for periodic scheduling problems with harmonic periods. In Proc. ICORES . SCITEPRESS, 2024. https://doi.org/10.5220/0012325800003639 doi:10.5220/0012325800003639

    work page doi:10.5220/0012325800003639 2024

  19. [19]

    u nzel, Harun Teper, Georg von der Br\

    Mario G\" u nzel, Harun Teper, Georg von der Br\" u ggen, and Jian-Jia Chen. End-to-end latency of cause-effect chains: A tutorial. ACM Trans. Embed. Comput. Syst. , 24(1), December 2024. https://doi.org/10.1145/3703630 doi:10.1145/3703630

    work page doi:10.1145/3703630 2024

  20. [20]

    When machine learning and neural networks marry real-time scheduling

    Zhishan Guo. When machine learning and neural networks marry real-time scheduling. Real-Time Systems , 61(2):320--325, 2025. https://doi.org/10.1007/s11241-025-09451-x doi:10.1007/s11241-025-09451-x

    work page doi:10.1007/s11241-025-09451-x 2025

  21. [21]

    Optimal task phasing for end-to-end latency in harmonic and semi-harmonic automotive systems

    Mario Günzel and Matthias Becker. Optimal task phasing for end-to-end latency in harmonic and semi-harmonic automotive systems. In Proc. RTAS , 2025. https://doi.org/10.1109/RTAS65571.2025.00026 doi:10.1109/RTAS65571.2025.00026

    doi:10.1109/rtas65571.2025.00026 2025

  22. [22]

    Communication Centric Design in Complex Automotive Embedded Systems

    Arne Hamann, Dakshina Dasari, Simon Kramer, Michael Pressler, and Falk Wurst. Communication Centric Design in Complex Automotive Embedded Systems . In Proc. ECRTS , volume 76, 2017. https://doi.org/10.4230/LIPIcs.ECRTS.2017.10 doi:10.4230/LIPIcs.ECRTS.2017.10

    work page doi:10.4230/lipics.ecrts.2017.10 2017

  23. [23]

    Reinforcement learning for search tree size minimization in constraint programming: New results on scheduling benchmarks

    Vilém Heinz, Petr Vilím, and Zdeněk Hanzálek. Reinforcement learning for search tree size minimization in constraint programming: New results on scheduling benchmarks. Computers & Industrial Engineering , 209:111413, 2025. https://doi.org/10.1016/j.cie.2025.111413 doi:10.1016/j.cie.2025.111413

    work page doi:10.1016/j.cie.2025.111413 2025

  24. [24]

    Isovic and G

    D. Isovic and G. Fohler. Efficient scheduling of sporadic, aperiodic, and periodic tasks with complex constraints. In Proc. RTSS , 2000. https://doi.org/10.1109/REAL.2000.896010 doi:10.1109/REAL.2000.896010

    work page doi:10.1109/real.2000.896010 2000

  25. [25]

    Kirsch and Ana Sokolova

    Christoph M. Kirsch and Ana Sokolova. The Logical Execution Time Paradigm . Springer Berlin Heidelberg, Berlin, Heidelberg, 2012. https://doi.org/10.1007/978-3-642-24349-3_5 doi:10.1007/978-3-642-24349-3_5

    work page doi:10.1007/978-3-642-24349-3_5 2012

  26. [26]

    Constrained data-age with job-level dependencies: How to reconcile tight bounds and overheads

    Tobias Klaus, Matthias Becker, Wolfgang Schröder-Preikschat, and Peter Ulbrich. Constrained data-age with job-level dependencies: How to reconcile tight bounds and overheads. In Proc. RTAS , 2021. https://doi.org/10.1109/RTAS52030.2021.00014 doi:10.1109/RTAS52030.2021.00014

    doi:10.1109/rtas52030.2021.00014 2021

  27. [27]

    Kramer, D

    S. Kramer, D. Ziegenbein, and A. Hamann. Real world automotive benchmarks for free. In Proc. WATERS , 2015

    2015

  28. [28]

    C. L. Liu and James W. Layland. Scheduling algorithms for multiprogramming in a hard-real-time environment. J. ACM , 20(1), 1973. https://doi.org/10.1145/321738.321743 doi:10.1145/321738.321743

    doi:10.1145/321738.321743 1973

  29. [29]

    McLean, Silviu S

    Shane D. McLean, Silviu S. Craciunas, Emil Alexander Juul Hansen, and Paul Pop. Mapping and scheduling automotive applications on ADAS platforms using metaheuristics. In Proc. ETFA , volume 1, pages 329--336, 2020. https://doi.org/10.1109/ETFA46521.2020.9212029 doi:10.1109/ETFA46521.2020.9212029

    doi:10.1109/etfa46521.2020.9212029 2020

  30. [30]

    Optimal harmonic period assignment: complexity results and approximation algorithms

    Morteza Mohaqeqi, Mitra Nasri, Yang Xu, Anton Cervin, and Karl-Erik A rz\' e n. Optimal harmonic period assignment: complexity results and approximation algorithms. Real-Time Syst. , 54(4), 2018

    2018

  31. [31]

    Guest editorial: a roadmap towards learning-enabled and learning-assisted real-time systems

    Mitra Nasri and Sanjoy Baruah. Guest editorial: a roadmap towards learning-enabled and learning-assisted real-time systems. Real-Time Systems , 61(2):183--184, 2025. https://doi.org/10.1007/s11241-025-09453-9 doi:10.1007/s11241-025-09453-9

    work page doi:10.1007/s11241-025-09453-9 2025

  32. [32]

    An efficient method for assigning harmonic periods to hard real-time tasks with period ranges

    Mitra Nasri and Gerhard Fohler. An efficient method for assigning harmonic periods to hard real-time tasks with period ranges. In Proc. ECRTS , 2015. https://doi.org/10.1109/ECRTS.2015.21 doi:10.1109/ECRTS.2015.21

    work page doi:10.1109/ecrts.2015.21 2015

  33. [33]

    A framework to construct customized harmonic periods for real-time systems

    Mitra Nasri, Gerhard Fohler, and Mehdi Kargahi. A framework to construct customized harmonic periods for real-time systems. In Proc. ECRTS , 2014. https://doi.org/10.1109/ECRTS.2014.31 doi:10.1109/ECRTS.2014.31

    work page doi:10.1109/ecrts.2014.31 2014

  34. [34]

    Deterministic architecture and middleware for domain control units and simplified integration process applied to ADAS

    Georg Niedrist. Deterministic architecture and middleware for domain control units and simplified integration process applied to ADAS . In Fahrerassistenzsysteme 2016 . Springer Fachmedien Wiesbaden, 2018. https://doi.org/10.1007/978-3-658-21444-9 doi:10.1007/978-3-658-21444-9

    work page doi:10.1007/978-3-658-21444-9 2016

  35. [35]

    Feasibility analysis of real-time periodic tasks with offsets

    Rodolfo Pellizzoni and Giuseppe Lipari. Feasibility analysis of real-time periodic tasks with offsets. Real-Time Syst. , 30(1–2), 2005. https://doi.org/10.1007/s11241-005-0506-x doi:10.1007/s11241-005-0506-x

    work page doi:10.1007/s11241-005-0506-x 2005

  36. [36]

    Data-age analysis and optimisation for cause-effect chains in automotive control systems

    Johannes Schlatow, Mischa Mostl, Sebastian Tobuschat, Tasuku Ishigooka, and Rolf Ernst. Data-age analysis and optimisation for cause-effect chains in automotive control systems. In Proc. SIES , 2018. https://doi.org/10.1109/SIES.2018.8442077 doi:10.1109/SIES.2018.8442077

    work page doi:10.1109/sies.2018.8442077 2018

  37. [37]

    Smit, Jianan Zhou, Robbert Reijnen, Yaoxin Wu, Jian Chen, Cong Zhang, Zaharah Bukhsh, Yingqian Zhang, and Wim Nuijten

    Igor G. Smit, Jianan Zhou, Robbert Reijnen, Yaoxin Wu, Jian Chen, Cong Zhang, Zaharah Bukhsh, Yingqian Zhang, and Wim Nuijten. Graph neural networks for job shop scheduling problems: A survey. Computers & Operations Research , 176:106914, 2025. https://doi.org/10.1016/j.cor.2024.106914 doi:10.1016/j.cor.2024.106914

    doi:10.1016/j.cor.2024.106914 2025

  38. [38]

    Spuri and J.A

    M. Spuri and J.A. Stankovic. How to integrate precedence constraints and shared resources in real-time scheduling. IEEE Transactions on Computers , 43(12):1407--1412, 1994. https://doi.org/10.1109/12.338100 doi:10.1109/12.338100

    work page doi:10.1109/12.338100 1994

  39. [39]

    Intelligent task scheduling for multi-core processors based on graph neural networks and deep reinforcement learning

    Yibo Zhou, Huabiao Qin, and Guancheng Chen. Intelligent task scheduling for multi-core processors based on graph neural networks and deep reinforcement learning. In Proc. ICAICE , page 875–880, New York, NY, USA, 2024. Association for Computing Machinery. https://doi.org/10.1145/3652628.3652774 doi:10.1145/3652628.3652774

    work page doi:10.1145/3652628.3652774 2024