Pith. sign in

REVIEW 4 major objections 5 minor 66 references

PromiseTune: Unveiling Causally Promising and Explainable Configuration Tuning

T0 review · 4 major / 5 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 PromiseTune claims that tuning guided by causally purified rules, which approximate promising regions of the configuration landscape, outperforms 11 state-of-the-art tuners by a 42% superior rank while providing spatial explainability of…

desk verdict Strong engineering result with an over-claimed causal core: the tuning gains look real, but the paper does not show that causal purification, rather than rule-bounded sampling, earns them. read the letter →

arxiv 2507.05995 v4 pith:MLG6OYPJ submitted 2025-07-08 cs.SE

Pengzhou Chen , Tao Chen This is my paper · ORCID
classification cs.SE
keywords configurationtuningcausalinferencerulelearningBayesianoptimizationexplainabilitylandscapesoftwareperformancesearch-basedengineering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes PromiseTune, a configuration tuner that learns rules from a Random Forest, purifies them with causal inference, and uses the surviving rules to bound where the tuner searches. The central claim is that these causally purified rules approximate the promising regions of a system's configuration landscape, so guiding a Bayesian optimizer with them mitigates the exploration-exploitation trade-off and yields better, more stable performance than existing tuners. The authors report that PromiseTune achieves 42% better rank than the overall second-best tuner across 12 systems and four budgets, and that the surviving rules offer fine-grained spatial explanations of which option ranges and interactions produce good configurations. A reader should care because configuration tuning is a costly bottleneck for modern software, and a tuner that can identify why certain regions are promising makes both tuning and future system design more efficient.

What carries the argument

The central mechanism is the causally purified rule: each rule is a path extracted from a Random Forest trained on measured configurations, bounding a region of the configuration landscape; each measured configuration is then featurized as a binary vector indicating which rules it fits. A causal graph is learned over these rule-membership features and the performance metric with Fast Causal Inference, and a rule survives purification only if it lies on a path to performance and has a negative average causal effect $\theta(p,R_i) < 0$, computed in do-calculus as the difference between the mean performance of configurations that fit and configurations that violate the rule. The surviving rules define where the optimizer samples, and the whole loop—rule learning, causal purification, and the Random Forest performance model—is re-run after every new measurement.

What would settle it

Measure PromiseTune on a synthetic configuration landscape with a known promising region and an explicit ground-truth performance function; if the purified rules fail to cover that region, or if PromiseTune does not beat its own no-rules ablation on the same landscape, the central claim is refuted.

Watch

Extended reading notes

Core claim

On the authors' own terms, the discovery is that rules read off the paths of a Random Forest, featurized as binary memberships and then purified by a causal graph built with Fast Causal Inference plus an average-causal-effect threshold, capture enough of the promising regions to guide a model-based optimizer toward better configurations. The retained rules with negative causal effect on the performance metric serve as spatial approximations of promising regions; they bound the sampling in a Bayesian optimization loop that uses Expected Improvement, and they are updated together with the surrogate model as new measurements arrive. The paper claims this yields significantly better and more stable performance than 11 state-of-the-art tuners, with the best or second-best Scott-Knott rank in 45 of 48 cases, and at the same time turns the final rules into spatially explainable output: option importance, option interactions, and the most plausible promising regions.

Load-bearing premise

The method's success depends on the assumption that, with as few as ten initial measurements, the causal graph and the group-mean difference in performance between rule-fitting and rule-violating configurations correctly identify which rules bound the genuinely good regions of the configuration space.

Editorial extensions

If this is right

  • Approximating promising regions with purified rules reduces the impact of the exploration-exploitation trade-off, which is the paper's stated mechanism for the performance gains.
  • On 12 systems across four budgets, PromiseTune is ranked best or second-best in 45 of 48 cases, with an overall rank 42% better than the second-best tuner.
  • The final purified rules, after fitting to the top $k\%$ measured configurations, provide spatial explainability—option importance, interactions, and most likely promising regions—beyond what option-level explainable tuners give.

Reading between the lines

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

  • Because the surviving rules describe regions rather than single points, they could plausibly transfer across software versions or workloads as reusable system-specific knowledge, a step the paper does not take.
  • A natural sensitivity check would replace the unadjusted group-mean difference in Equation 4 with a covariate-adjusted estimate; if the surviving rule set changes materially, the guidance depends on that estimator choice.
  • The same causal purification could be applied to guide multi-objective tuning or to warm-start other optimizers, which the paper leaves implicit.
  • The rule-based explanation could be tested against expert annotations of which regions truly are promising, going beyond the case study in the paper.
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

4 major / 5 minor

Summary. The paper proposes PromiseTune, a configuration tuner that learns rules from a Random Forest, featurizes configurations by rule membership, purifies rules via Fast Causal Inference (FCI) and an average causal effect threshold, and then uses the surviving rules to bound the sampling region of a Random-Forest-based Bayesian optimizer. The authors claim that this causally purified rule guidance mitigates the exploration/exploitation trade-off and yields significantly better and more stable performance than 11 existing tuners across 12 systems and four budgets, while also providing spatial explainability of promising regions. The evaluation consists of 48 comparison cases with 30 runs each, Scott-Knott ESD ranking, sensitivity analyses for parameters l and k, and a qualitative explainability case study on x264.

Significance. If the central claims hold, PromiseTune would be a practically useful contribution to configuration tuning: it combines a concrete mechanism for focusing search on promising regions with a novel landscape-level explanation output, and the empirical study is unusually broad, covering 12 diverse systems and four budget levels with 30 runs per case. The paper also ships its source code and raw data, which supports reproducibility. However, the causal attribution at the heart of the method is not currently established: the do-calculus effect in Equation (4) is implemented as an unadjusted group-mean difference, and the RQ2 ablation cannot separate causal purification from rule-guided sampling in general. Because the paper's novelty and the 'solely contributed by the causally-purified rules' claim rest on this causal mechanism, these issues are load-bearing and need to be addressed before the contribution can be accepted as stated.

major comments (4)
  1. [Section 5.2, footnote 5, Figure 6] The ablation in RQ2 does not isolate causal purification, so the claim that the gains are 'solely contributed by the causally-purified rules' is not supported. The w/o Rules variant disables both the rule generator and the purifier, and footnote 5 states that using all learned rules is equivalent to sampling the entire configuration landscape; thus Figure 6 compares PromiseTune against a tuner with no rule guidance at all, not against a rule-guided tuner without the causal filter. Please add a control condition that keeps rule-guided optimization fixed while removing only the FCI/θ purification step (for example, using randomly selected or frequency-selected subsets of learned rules of the same cardinality as the purified set), or soften the attribution claim accordingly.
  2. [Section 3.2.2, Equation (4)] The do-calculus estimand in Equation (4) is not what is actually computed. The text describes θ(p, R_i) as the difference between the expected performance of configurations that fit versus violate R_i in the featurized dataset of Equation (3), which is an unadjusted associational difference, not E[f|do(r_i=1)] − E[f|do(r_i=0)]. No back-door adjustment, instrumental variable, or other identification strategy is described, even though rule-membership features are derived from the same Random Forest and are mutually correlated. Please either implement a proper adjustment using the FCI graph or reframe the method as association-based purification and remove the do-calculus/causal-effect terminology from the load-bearing claims.
  3. [Section 3.2.1 and Algorithm 1] FCI is applied to very small samples (the initial sample size is s=10 in Algorithm 1, and even later iterations have at most a few hundred measured points) over a featurized space whose dimension equals the number of learned rules, which can be large. The stability of the resulting partial ancestral graph is not reported, and there is no validation that the graph's paths to the performance node are reliable; this matters because the graph determines which rules enter the θ computation in Equation (4). Please report graph stability across seeds and/or validate the discovered rules against known promising regions on at least a subset of the systems, for instance by comparing purified rules with the empirical top-performing regions on held-out evaluations.
  4. [Section 3.4, Section 5.4, Algorithm 1 line 22] The explainability result is partly tautological. The explainable rules ℛ′_p are defined as the purified rules that fit the top k% of the measured configurations (Algorithm 1, line 22), and RQ4 then reports that these rules cover the top configurations and that the most common overlaps bound good configurations. Because the selection criterion is the same as the evaluation criterion, this does not provide independent evidence that the rules reveal hidden landscape structure. Please add an out-of-sample or independent validation (for example, hold-out configurations not used in rule selection, or synthetic landscapes with known ground-truth promising regions) or reframe the claim as descriptive summarization of the measured data.
minor comments (5)
  1. [Section 6.1] The section title 'Why dose PromiseTune Work?' contains a typo and should read 'Why does PromiseTune Work?'.
  2. [Section 3.1.2] The sentence 'We featurizing the rules over all configurations' is ungrammatical; it should be 'We featurize the rules over all configurations'.
  3. [Algorithm 1, line 14] The loop condition 'while sample more for C′ can still improve αEI' is not formally defined; the GKDE-based stopping criterion is only described narratively in Section 3.3, which makes the pseudocode non-reproducible without cross-referencing the prose.
  4. [Section 5.1.2] The sentence 'The improvements of PromiseTune is overall significant' should be 'The improvements of PromiseTune are overall significant'.
  5. [Section 3.2.2] The notation switches between f in Equation (4) and p in the surrounding text (for example, θ(p, R_i)); please unify the notation for the performance variable.

Circularity Check

1 steps flagged · score 6.0 of 10

Explainability results are self-definitional: the 'promising regions' are selected to fit the top k% measured configurations and then verified against those same configurations.

  1. self definitional [Section 3.4 'Explainability with Purified Rules'; Algorithm 1 line 22; Section 5.4.2 RQ4]
    "by the end of the tuning, we use the measured configurations with top k% performance and examine which are the purified rules that those configurations fit. The ones that can be fitted, referred to as explainable rules, are then returned. ... We verified that the two most promising regions can indeed bound most of the good configurations."

    The explainable rules are, by construction, exactly those purified rules that fit the top k% of measured configurations (Algorithm 1 line 22). The RQ4 verification that the returned 'most promising regions' bound most good configurations is therefore an immediate consequence of the selection criterion rather than an independent confirmation. The paper presents this as evidence that PromiseTune 'provide[s] richer information to explain the hidden system characteristics,' but the spatial explanation is just a description of the locations of already-selected good configurations. No independent data is used; the output is defined in terms of the target it is claimed to explain.

full rationale

PromiseTune's core tuning loop is non-circular: rules are extracted from a Random Forest fitted to measured configurations, purified by an in-sample causal-effect screen, and then used to bound Bayesian-optimization sampling; the resulting performance is benchmarked against external tuners. The RQ2 ablation is not itself a circular reduction (removing all rule guidance is a confound, not a construction), and the self-citations for the promising-region premise are backed by Figure 2, so they are not load-bearing. However, the explainability claim contains a definitional circularity. Section 3.4 defines the returned 'explainable rules' as those purified rules that fit the top k% of measured configurations, and RQ4 then treats the fact that these rules bound good configurations as a discovered verification. That verification is true by construction: the rules were chosen to fit those same configurations. The spatial explanation is therefore a restatement of the selection criterion rather than an independent finding. Because the paper presents this explainability as a central contribution ('providing richer information to explain the hidden system characteristics'), the circularity is material, though it does not infect the measured performance comparison. Score 6 reflects one prediction-like output reducing by construction.

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

No new physical or conceptual entities are introduced. The approach relies on standard tools (random forests, FCI, Bayesian optimization) plus three domain assumptions about the configuration landscape and the validity of the causal estimate.

free parameters (2)
  • l (minimal leaf count for random forest) = 10 (default)
    Controls the number of rules learned; sensitivity is studied in RQ3, so it is chosen by hand rather than fitted to data.
  • k (top k% configurations for explainable rules) = varies, e.g., 10 in the case study
    Only affects the explanation output, not the optimization; the paper states it is case-dependent.
assumptions (3)
  • domain assumption Most good configurations condense into a small number of promising regions.
    Motivates the entire approach; cited from prior work (FLASH [44], Chen et al. [9,10]) but treated as a premise.
  • domain assumption FCI causal discovery on rule-membership features yields a valid causal graph under small samples.
    Section 3.2.1 applies FCI to a few dozen to a few hundred samples with many rule features; no faithfulness or sufficiency conditions are discussed.
  • ad hoc to paper The average causal effect theta(p, R_i) in Equation 4 is identified by the difference between fitted and violated configurations.
    Section 3.2.2 computes group-mean differences and labels them do-calculus effects; no back-door adjustment or other identification strategy is specified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PromiseTune: Unveiling Causally Promising and Explainable Configuration Tuning." pith.science (2026). https://pith.science/paper/MLG6OYPJ

@misc{pith2026250705995,
  author       = {Pith},
  title        = {Pith review of: PromiseTune: Unveiling Causally Promising and Explainable Configuration Tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MLG6OYPJ}},
  note         = {Machine review of arXiv:2507.05995}
}
read the original abstract

The high configurability of modern software systems has made configuration tuning a crucial step for assuring system performance, e.g., latency or throughput. However, given the expensive measurements, large configuration space, and rugged configuration landscape, existing tuners suffer ineffectiveness due to the difficult balance of budget utilization between exploring uncertain regions (for escaping from local optima) and exploiting guidance of known good configurations (for fast convergence). The root cause is that we lack knowledge of where the promising regions lay, which also causes challenges in the explainability of the results. In this paper, we propose PromiseTune that tunes configuration guided by causally purified rules. PromiseTune is unique in the sense that we learn rules, which reflect certain regions in the configuration landscape, and purify them with causal inference. The remaining rules serve as approximated reflections of the promising regions, bounding the tuning to emphasize these places in the landscape. This, as we demonstrate, can effectively mitigate the impact of the exploration and exploitation trade-off. Those purified regions can then be paired with the measured configurations to provide spatial explainability at the landscape level. Comparing with 11 state-of-the-art tuners on 12 systems and varying budgets, we show that PromiseTune performs significantly better than the others with 42% superior rank to the overall second best while providing richer information to explain the hidden system characteristics.

Figures

Figures reproduced from arXiv: 2507.05995 by the authors.

Figure 1
Figure 1. Example of Redis system. (a) is the projected config￾uration landscape; (b) is the tuning trajectories of two tuners. ACM Reference Format: Pengzhou Chen and Tao Chen. 2026. PromiseTune: Unveiling Causally Promising and Explainable Configuration Tuning. In 2026 IEEE/ACM 48th International Conference on Software Engineering (ICSE ’26), April 12–18, 2026, Rio de Janeiro, Brazil. ACM, New York, NY, USA, 13 pages. https… view at source ↗
Figure 2
Figure 2. Projection of the configuration landscape for two [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Workflow overview of PromiseTune. measurements to focus on the best region found so far might stick at local optima forever [8, 27, 44]. To understand the root causes, we analyze the landscape of configurable systems [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Example of purification via causal inference. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Ablating causally purified rules over 30 runs (smaller performance is better). [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Sensitivity of PromiseTune to parameter 𝑙 over all systems. The smaller the normalized performance, the better. Those demonstrate that, regardless of the systems or budgets, the rules, after purification, can effectively guide the tuning towards the promising regions, …
Figure 8
Figure 8. Figure 8: The explored configurations within the last [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 23 canonical work pages

  1. [1]

    Gordon, and Bohan Zhang

    Dana Van Aken, Andrew Pavlo, Geoffrey J. Gordon, and Bohan Zhang. 2017. Automatic Database Management System Tuning Through Large-scale Machine Learning. In Proceedings of the 2017 ACM International Conference on Management of Data, SIGMOD Conference 2017, Chicago, IL, USA, May 14-19, 2017 . 1009–1024. doi:10.1145/3035918.3064029

    arXiv 2017

  2. [2]

    Awad, Neeratyoy Mallik, and Frank Hutter

    Noor H. Awad, Neeratyoy Mallik, and Frank Hutter. 2021. DEHB: Evolutionary Hyberband for Scalable, Robust and Efficient Hyperparameter Optimization. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI 2021, Virtual Event / Montreal, Canada, 19-27 August 2021 , Zhi-Hua Zhou (Ed.). ijcai.org, 2147–2153. doi:10.24...

    doi:10.24963/ijcai.2021/296 2021

  3. [3]

    Aydt, Quincey Koziol, and Marc Snir

    Babak Behzad, Huong Vu Thanh Luu, Joseph Huchette, Surendra Byna, Prabhat, Ruth A. Aydt, Quincey Koziol, and Marc Snir. 2013. Taming parallel I/O complexity with auto-tuning. In International Conference for High Performance Computing, Networking, Storage and Analysis, SC’13, Denver, CO, USA - November 17 - 21, 2013, William Gropp and Satoshi Matsuoka (Eds...

    arXiv 2013

  4. [4]

    Leo Breiman. 2001. Random forests. Machine learning 45 (2001), 5–32

    2001

  5. [5]

    Baoqing Cai, Yu Liu, Ce Zhang, Guangyu Zhang, Ke Zhou, Li Liu, Chunhua Li, Bin Cheng, Jie Yang, and Jiashu Xing. 2022. HUNTER: An Online Cloud Database Hybrid Tuning System for Personalized Requirements. In SIGMOD ’22: International Conference on Management of Data, Philadelphia, PA, USA, June 12 - 17, 2022. 646–659. doi:10.1145/3514221.3517882

    arXiv 2022

  6. [6]

    Wu, Panpan Zhangsun, Yufei Li, and Zhe Zhang

    Rong Cao, Liang Bao, Chase Q. Wu, Panpan Zhangsun, Yufei Li, and Zhe Zhang

    work page

  7. [7]

    Sampling

    Jianfeng Chen, Vivek Nair, Rahul Krishna, and Tim Menzies. 2019. "Sampling" as a Baseline Optimizer for Search-Based Software Engineering. IEEE Trans. Software Eng. 45, 6 (2019), 597–614. doi:10.1109/TSE.2018.2790925

    arXiv 2019

  8. [9]

    Pengzhou Chen, Tao Chen, and Miqing Li. 2024. MMO: Meta Multi- Objectivization for Software Configuration Tuning. IEEE Trans. Software Eng. 50, 6 (2024), 1478–1504. doi:10.1109/TSE.2024.3388910

    arXiv 2024

Show all 66 references
  1. [10]

    Pengzhou Chen, Jingzhi Gong, and Tao Chen. 2025. Accuracy Can Lie: On the Impact of Surrogate Model in Configuration Tuning. IEEE Transactions on Software Engineering 51, 2 (2025), 548–580. doi:10.1109/TSE.2025.3525955

    2025

  2. [11]

    Tao Chen. 2022. Lifelong Dynamic Optimization for Self-Adaptive Systems: Fact or Fiction?. In IEEE International Conference on Software Analysis, Evolution and Reengineering, SANER 2022, Honolulu, HI, USA, March 15-18, 2022 . IEEE, 78–89. doi:10.1109/SANER53432.2022.00022

    2022

  3. [12]

    Tao Chen and Rami Bahsoon. 2017. Self-Adaptive and Online QoS Modeling for Cloud-Based Software Services. IEEE Trans. Software Eng. 43, 5 (2017), 453–475. doi:10.1109/TSE.2016.2608826

    2017

  4. [13]

    Tao Chen and Rami Bahsoon. 2017. Self-Adaptive Trade-off Decision Making for Autoscaling Cloud-Based Services. IEEE Trans. Serv. Comput. 10, 4 (2017), 618–632. doi:10.1109/TSC.2015.2499770

    2017

  5. [14]

    Tao Chen, Ke Li, Rami Bahsoon, and Xin Yao. 2018. FEMOSAA: Feature-Guided and Knee-Driven Multi-Objective Optimization for Self-Adaptive Software. ACM Trans. Softw. Eng. Methodol. 27, 2 (2018), 5:1–5:50. doi:10.1145/3204459

    2018 doi

  6. [15]

    Tao Chen and Miqing Li. 2021. Multi-objectivizing software configuration tuning. In ESEC/FSE ’21: 29th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, Athens, Greece, August 23-28, 2021. 453–465. doi:10.1145/3468264.3468555

    2021

  7. [16]

    Tao Chen and Miqing Li. 2023. Do Performance Aspirations Matter for Guiding Software Configuration Tuning? An Empirical Investigation under Dual Perfor- mance Objectives. ACM Trans. Softw. Eng. Methodol. 32, 3 (2023), 68:1–68:41. doi:10.1145/3571853

    2023 doi

  8. [17]

    Tao Chen and Miqing Li. 2023. The Weights Can Be Harmful: Pareto Search versus Weighted Search in Multi-objective Search-based Software Engineering. ACM Trans. Softw. Eng. Methodol. 32, 1 (2023), 5:1–5:40. doi:10.1145/3514233

    2023 doi

  9. [18]

    Tao Chen and Miqing Li. 2024. Adapting Multi-objectivized Software Configura- tion Tuning. Proc. ACM Softw. Eng. 1, FSE (2024), 539–561. doi:10.1145/3643751

    2024 doi

  10. [19]

    Alexander I Cowen-Rivers, Wenlong Lyu, Rasul Tutunov, Zhi Wang, Antoine Grosnit, Ryan Rhys Griffiths, Alexandre Max Maraval, Hao Jianye, Jun Wang, Jan Peters, et al. 2022. Hebo: Pushing the limits of sample-efficient hyper-parameter optimisation. Journal of Artificial Intellig...

    2022

  11. [20]

    Baljinder Ghotra, Shane McIntosh, and Ahmed E. Hassan. 2015. Revisiting the Impact of Classification Techniques on the Performance of Defect Prediction Models. In 37th IEEE/ACM International Conference on Software Engineering, ICSE 2015, Florence, Italy, May 16-24, 2015, Volum...

    2015

  12. [21]

    Jingzhi Gong and Tao Chen. 2023. Predicting Software Performance with Divide- and-Learn. In Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, ESEC/FSE 2023, San Francisco, CA, USA, December 3-9,...

    2023 doi

  13. [22]

    Jingzhi Gong and Tao Chen. 2024. Predicting Configuration Performance in Multiple Environments with Sequential Meta-Learning. Proceedings of ACM Software Engineering 1, FSE (2024), 359–382. doi:10.1145/3643743

    2024 doi

  14. [23]

    Jingzhi Gong and Tao Chen. 2025. Deep Configuration Performance Learning: A Systematic Survey and Taxonomy. ACM Trans. Softw. Eng. Methodol. 34, 1 (2025), 25:1–25:62. doi:10.1145/3702986

    2025 doi

  15. [24]

    Jingzhi Gong, Tao Chen, and Rami Bahsoon. 2025. Dividable Configuration Performance Learning. IEEE Trans. Software Eng. 51, 1 (2025), 106–134. doi:10. 1109/TSE.2024.3491945

    2025

  16. [25]

    An Empirical Comparison of Model Validation Techniques for Defect Prediction Models

    Steffen Herbold. 2017. Comments on ScottKnottESD in Response to "An Empirical Comparison of Model Validation Techniques for Defect Prediction Models". IEEE Trans. Software Eng. 43, 11 (2017), 1091–1094. doi:10.1109/TSE.2017.2748129

    2017

  17. [26]

    Abir Hossen, Sonam Kharade, Jason M

    Md. Abir Hossen, Sonam Kharade, Jason M. O’Kane, Bradley R. Schmerl, David Garlan, and Pooyan Jamshidi. 2024. CURE: Simulation-Augmented Auto-Tuning in Robotics. CoRR abs/2402.05399 (2024). doi:10.48550/ARXIV.2402.05399 arXiv:2402.05399

    2024 doi arXiv

  18. [27]

    Frank Hutter, Holger H Hoos, and Kevin Leyton-Brown. 2011. Sequential model- based optimization for general algorithm configuration. InLearning and Intelligent Optimization: 5th International Conference, LION 5, Rome, Italy, January 17-21,

    2011

  19. [28]

    Hoos, and Kevin Leyton-Brown

    Frank Hutter, Holger H. Hoos, and Kevin Leyton-Brown. 2011. Sequential Model- Based Optimization for General Algorithm Configuration. In Learning and Intel- ligent Optimization - 5th International Conference, LION 5, Rome, Italy, January 17-21, 2011. Selected Papers . 507–523....

    2011 doi

  20. [29]

    Md Shahriar Iqbal, Rahul Krishna, Mohammad Ali Javidian, Baishakhi Ray, and Pooyan Jamshidi. 2022. Unicorn: reasoning about configurable system performance through the lens of causality. In EuroSys ’22: Seventeenth Euro- pean Conference on Computer Systems, Rennes, France, Apr...

    2022

  21. [30]

    Md Shahriar Iqbal, Ziyuan Zhong, Iftakhar Ahmad, Baishakhi Ray, and Pooyan Jamshidi. 2023. CAMEO: A Causal Transfer Learning Approach for Performance Optimization of Configurable Computer Systems. In Proceedings of the 2023 ACM Symposium on Cloud Computing, SoCC 2023, Santa Cr...

    2023

  22. [31]

    Pooyan Jamshidi and Giuliano Casale. 2016. An Uncertainty-Aware Approach to Optimal Configuration of Stream Processing Systems. In 24th IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommuni- cation Systems, MASCOTS 2016, London, United K...

    2016

  23. [32]

    Pooyan Jamshidi, Miguel Velez, Christian Kästner, and Norbert Siegmund. 2018. Learning to sample: exploiting similarities across environments to learn perfor- mance models for configurable systems. In Proceedings of the 2018 ACM Joint Meeting on European Software Engineering C...

    2018

  24. [33]

    Konstantinos Kanellis, Cong Ding, Brian Kroth, Andreas Müller, Carlo Curino, and Shivaram Venkataraman. 2022. LlamaTune: Sample-Efficient DBMS Configu- ration Tuning. Proc. VLDB Endow. 15, 11 (2022), 2953–2965. doi:10.14778/3551793. 3551844

    2022 doi

  25. [34]

    Rahul Krishna and Tim Menzies. 2019. Bellwethers: A Baseline Method for Transfer Learning. IEEE Transactions on Software Engineering 45, 11 (2019), 1081–1105. doi:10.1109/TSE.2018.2821670

    2019

  26. [35]

    Rahul Krishna, Vivek Nair, Pooyan Jamshidi, and Tim Menzies. 2021. Whence to Learn? Transferring Knowledge in Configurable Systems Using BEETLE. IEEE Trans. Software Eng. 47, 12 (2021), 2956–2972. doi:10.1109/TSE.2020.2983927

    2021

  27. [36]

    Jiale Lao, Yibo Wang, Yufei Li, Jianping Wang, Yunjia Zhang, Zhiyuan Cheng, Wanghu Chen, Mingjie Tang, and Jianguo Wang. 2024. GPTuner: A Manual- Reading Database Tuning System via GPT-Guided Bayesian Optimization. Proc. VLDB Endow. 17, 8 (2024), 1939–1952. doi:10.14778/365943...

    2024

  28. [37]

    Yiyan Li, Haoyang Li, Pu Zhao, Jing Zhang, Xinyi Zhang, Tao Ji, Luming Sun, Cuiping Li, and Hong Chen. 2024. Is Large Language Model Good at Database Knob Tuning? A Comprehensive Experimental Evaluation. CoRR abs/2408.02213 (2024). doi:10.48550/ARXIV.2408.02213 arXiv:2408.02213

    2024 doi arXiv

  29. [38]

    Yang Li, Yu Shen, Wentao Zhang, Yuanwei Chen, Huaijun Jiang, Mingchao Liu, Jiawei Jiang, Jinyang Gao, Wentao Wu, Zhi Yang, Ce Zhang, and Bin Cui. 2021. OpenBox: A Generalized Black-box Optimization Service. In KDD ’21: The 27th ACM SIGKDD Conference on Knowledge Discovery and ...

    2021

  30. [41]

    Patrick E McKight and Julius Najab. 2010. Kruskal-wallis test. The corsini encyclopedia of psychology (2010), 1–1

    2010

  31. [42]

    Nikolaos Mittas and Lefteris Angelis. 2013. Ranking and Clustering Software Cost Estimation Models through a Multiple Comparisons Algorithm. IEEE Trans. Software Eng. 39, 4 (2013), 537–551. doi:10.1109/TSE.2012.45

    2013 doi

  32. [44]

    Vivek Nair, Zhe Yu, Tim Menzies, Norbert Siegmund, and Sven Apel. 2020. Finding Faster Configurations Using FLASH. IEEE Trans. Software Eng. 46, 7 (2020), 794–

    2020

  33. [45]

    Judea Pearl et al. 2000. Models, reasoning and inference. Cambridge, UK: Cam- bridgeUniversityPress 19, 2 (2000), 3

    2000

  34. [46]

    Mohammed Sayagh, Noureddine Kerzazi, Bram Adams, and Fábio Petrillo. 2020. Software Configuration Engineering in Practice Interviews, Survey, and Sys- tematic Literature Review. IEEE Trans. Software Eng. 46, 6 (2020), 646–673. doi:10.1109/TSE.2018.2867847

    2020

  35. [47]

    Arman Shahbazian, Suhrid Karthik, Yuriy Brun, and Nenad Medvidovic. 2020. eQual: informing early design decisions. InESEC/FSE ’20: 28th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, Virtual Event, USA, November 8-1...

    2020 doi

  36. [48]

    Shohei Shimizu. 2014. LiNGAM: Non-Gaussian methods for estimating causal structures. Behaviormetrika 41, 1 (2014), 65–98

    2014

  37. [49]

    Peter Spirtes, Clark Glymour, and Richard Scheines. 2001. Causation, prediction, and search. MIT press

    2001

  38. [50]

    Richardson

    Peter Spirtes, Christopher Meek, and Thomas S. Richardson. 1995. Causal Inference in the Presence of Latent Variables and Selection Bias. In UAI ’95: Proceedings of the Eleventh Annual Conference on Uncertainty in Ar- tificial Intelligence, Montreal, Quebec, Canada, August 18-...

    1995

  39. [51]

    Sullivan, Margo I

    David G. Sullivan, Margo I. Seltzer, and Avi Pfeffer. 2004. Using probabilistic reasoning to automate software tuning. In Proceedings of the International Con- ference on Measurements and Modeling of Computer Systems, SIGMETRICS 2004, June 10-14, 2004, New York, NY, USA . 404–...

    2004

  40. [52]

    Hassan, and Kenichi Matsumoto

    Chakkrit Tantithamthavorn, Shane McIntosh, Ahmed E. Hassan, and Kenichi Matsumoto. 2019. The Impact of Automated Parameter Optimization on Defect Prediction Models. IEEE Trans. Software Eng. 45, 7 (2019), 683–711. doi:10.1109/ TSE.2018.2794977

    2019

  41. [53]

    George R Terrell and David W Scott. 1992. Variable kernel density estimation. The Annals of Statistics (1992), 1236–1265

    1992

  42. [54]

    Pavel Valov, Jean-Christophe Petkovich, Jianmei Guo, Sebastian Fischmeister, and Krzysztof Czarnecki. 2017. Transferring Performance Prediction Models Across Different Hardware Platforms. In Proceedings of the 8th ACM/SPEC on International Conference on Performance Engineering...

    2017

  43. [55]

    Miguel Velez, Pooyan Jamshidi, Norbert Siegmund, Sven Apel, and Christian Kästner. 2021. White-Box Analysis over Machine Learning: Modeling Perfor- mance of Configurable Systems. In 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE) . 1072–1084. doi:10....

    2021

  44. [56]

    Max Weber, Christian Kaltenecker, Florian Sattler, Sven Apel, and Norbert Sieg- mund. 2023. Twins or False Friends? A Study on Energy Consumption and Performance of Configurable Software. In 45th IEEE/ACM International Confer- ence on Software Engineering, ICSE 2023, Melbourne...

    2023

  45. [57]

    Zezhen Xiang, Jingzhi Gong, and Tao Chen. 2026. Dually Hierarchical Drift Adaptation for Online Configuration Performance Learning. In 48th IEEE/ACM International Conference on Software Engineering (ICSE) . ACM

    2026

  46. [58]

    Yulong Ye, Tao Chen, and Miqing Li. 2025. Distilled Lifelong Self-Adaptation for Configurable Systems. In 47th IEEE/ACM International Conference on Software Engineering, ICSE 2025, Ottawa, ON, Canada, April 26 - May 6, 2025 . IEEE, 1333–

    2025

  47. [59]

    Dawei Zhan and Huanlai Xing. 2020. Expected improvement for expensive optimization: a review. Journal of Global Optimization 78, 3 (2020), 507–544

    2020

  48. [60]

    Ji Zhang, Yu Liu, Ke Zhou, Guoliang Li, Zhili Xiao, Bin Cheng, Jiashu Xing, Yangtao Wang, Tianheng Cheng, Li Liu, Minwei Ran, and Zekang Li. 2019. An End-to-End Automatic Cloud Database Tuning System Using Deep Reinforcement Learning. In Proceedings of the 2019 International C...

    2019

  49. [61]

    Xinyi Zhang, Hong Wu, Zhuo Chang, Shuowei Jin, Jian Tan, Feifei Li, Tieying Zhang, and Bin Cui. 2021. ResTune: Resource Oriented Tuning Boosted by Meta-Learning for Cloud Databases. In SIGMOD ’21: International Conference on Management of Data, Virtual Event, China, June 20-25...

    2021

  50. [62]

    Mingxuan Zhu and Dan Hao. 2023. Compiler Auto-Tuning via Critical Flag Selection. In 38th IEEE/ACM International Conference on Automated Software Engineering, ASE 2023, Luxembourg, September 11-15, 2023 . 1000–1011. doi:10. 1109/ASE56229.2023.00209

    2023

  51. [63]

    Yuqing Zhu, Jianxun Liu, Mengying Guo, Yungang Bao, Wenlong Ma, Zhuoyue Liu, Kunpeng Song, and Yingchun Yang. 2017. BestConfig: tapping the perfor- mance potential of systems via automatic configuration tuning. In Proceedings of the 2017 Symposium on Cloud Computing, SoCC 2017...

    2017

  52. [64]

    Juliusz Krysztof Ziomek and Haitham Bou-Ammar. 2023. Are Random Decompo- sitions all we need in High Dimensional Bayesian Optimisation?. In International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA. 43347–43368. https://proceedings.mlr.pre...

    2023

  53. [506]

    https://dslpitt.org/uai/displayArticleDetails.jsp?mmnu=1&smnu=2&article_ id=469&proceeding_id=11

  54. [811]

    doi:10.1109/TSE.2018.2870895

    2018

  55. [1345]

    doi:10.1109/ICSE55347.2025.00094

    2025

  56. [2011]

    Springer, 507–523

    Selected Papers 5 . Springer, 507–523

  57. [2019]

    doi:10.1145/3299869.3300085

    415–432. doi:10.1145/3299869.3300085

  58. [2023]

    CM-CASL: Comparison-based performance modeling of software systems via collaborative active and semisupervised learning. J. Syst. Softw. 201 (2023), 111686. doi:10.1016/J.JSS.2023.111686

    2023

Pith tools

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