Pith. sign in

REVIEW 2 major objections 2 minor 1 cited by

PRIM performs root cause analysis by Bayesian inference over a synthetic prior of causal models, marginalizing structural uncertainty to detect changes between baseline and anomalous data.

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 ↗

PRIM meta-learns a Model-Averaged Causal Estimation transformer to perform Bayesian RCA by marginalizing structural uncertainty over synthetic causal priors, achieving 17ms inference on systems up to 100 variables.

T0 review reviewed 2026-06-30 challenge →

load-bearing objection PRIM's meta-learned RCA via synthetic causal priors is a distinct framing that delivers fast inference, but the load-bearing assumption on prior representativeness lacks visible checks. the 2 major comments →

arxiv 2605.08786 v3 pith:YDR4FNFU submitted 2026-05-09 cs.LG

PRIM: Meta-Learned Bayesian Root Cause Analysis

Christopher Lohse , Anish Dhir , Amadou Ba , Bradley Eck , Marco Ruffini , Jonas Wahl This is my paper
classification cs.LG
keywords root cause analysismeta-learningcausal inferenceBayesian inferenceneural processestransformer modelsanomaly detectionzero-shot inference
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.

The reading

The paper tries to establish that root cause analysis can be framed as Bayesian inference over many possible causal structures generated synthetically during training, allowing the method to identify which variables changed their generating mechanism without explicit statistical tests or fitting a graph at test time. A sympathetic reader would care because this removes the need to supply a causal graph or run slow inference on large systems, while still matching the accuracy of methods that do have the graph. The approach uses a transformer neural process to jointly attend to normal samples, anomalous samples, and node structures, producing inferences in 17 milliseconds even for systems with 100 variables. It remains competitive on synthetic benchmarks and two realistic datasets, with further gains from light domain-specific fine-tuning.

Core claim

PRIM frames RCA as a Bayesian inference task over a synthetic prior of causal models. By marginalising out structural uncertainty, it implicitly identifies changes in the data-generating mechanism between baseline and anomalous periods. It infers distributional differences without explicit statistical testing and implicitly learns causal structure without model fitting at test time. A Model-Averaged Causal Estimation transformer neural process jointly attends over observational and anomalous samples and the causal structure of nodes, enabling zero-shot inference in 17 ms for systems with up to 100 variables and remaining competitive with graph-aware methods on PetShop and CausRCA benchmarks.

What carries the argument

The Model-Averaged Causal Estimation (MACE) transformer neural process that jointly attends over observational and anomalous samples and the causal structure of nodes, marginalising structural uncertainty from the synthetic prior.

Load-bearing premise

The synthetic prior of causal models used in meta-training is representative enough of the structural uncertainty and data-generating processes in the target real systems.

What would settle it

On the PetShop or CausRCA benchmarks, measure whether PRIM's accuracy drops below that of graph-aware baselines by more than the reported margin even after fine-tuning, or whether inference time exceeds 17 ms for 100-variable systems.

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

If this is right

  • Root causes can be identified without supplying the system's causal graph at test time.
  • Inference completes in milliseconds for systems up to 100 variables.
  • Performance on specific domains improves with lightweight fine-tuning after meta-training.
  • The method remains competitive with approaches that receive the causal graph a priori.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same marginalisation strategy could be applied to other tasks that require implicit causal reasoning without an explicit graph.
  • If the synthetic prior can be made more realistic through additional simulation techniques, the gap to fully supervised causal methods might narrow further.
  • Testing on systems larger than 100 variables would show whether the transformer attention mechanism continues to scale without retraining.
Share X Bluesky LinkedIn Reddit HN

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 / 2 minor

Summary. The paper introduces PRIM, a causal meta-learning method for root cause analysis that frames RCA as Bayesian inference over a synthetic prior of causal models. Using a Model-Averaged Causal Estimation (MACE) transformer neural process, it marginalizes structural uncertainty to implicitly identify distributional changes and causal structure, enabling zero-shot inference in 17 ms for systems with up to 100 variables. It reports competitiveness with graph-aware baselines and outperformance of graph-unaware methods on synthetic benchmarks plus the PetShop and CausRCA real-world datasets, with further gains from lightweight fine-tuning.

Significance. If the meta-trained synthetic prior adequately represents the structural uncertainty and data-generating processes of target systems, the approach offers a notable advance in fast, graph-agnostic RCA for complex systems. The simulation-based meta-learning paradigm and joint attention over observational/anomalous samples plus causal structure are strengths that could support practical deployment where explicit causal graphs or per-instance fitting are unavailable.

major comments (2)
  1. [§5 (Experiments and Results)] The load-bearing assumption that the synthetic prior of causal models generated during meta-training is representative of PetShop and CausRCA is not supported by quantitative validation such as graph-property matching (e.g., degree distributions, density) or sensitivity analysis to prior misspecification; without this, the zero-shot competitiveness claim on real benchmarks risks reducing to benchmark-specific pattern matching rather than robust Bayesian marginalization.
  2. [§3.2] §3.2 (MACE transformer neural process): the claim of implicitly learning causal structure without model fitting at test time requires explicit confirmation that the attention mechanism over causal structure nodes does not implicitly rely on information leaked from the meta-training distribution that matches the test benchmarks; otherwise the 'parameter-free' zero-shot inference may not hold under distribution shift.
minor comments (2)
  1. The abstract contains a typographical error ('17,ms' instead of '17 ms').
  2. [§3] Notation for the MACE architecture and the marginalization over structural uncertainty could be clarified with an explicit equation for the posterior predictive in the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for stronger validation of the synthetic prior's representativeness and explicit safeguards against potential distribution leakage in the MACE transformer. We respond to each major comment below.

read point-by-point responses

  1. Referee: [§5 (Experiments and Results)] The load-bearing assumption that the synthetic prior of causal models generated during meta-training is representative of PetShop and CausRCA is not supported by quantitative validation such as graph-property matching (e.g., degree distributions, density) or sensitivity analysis to prior misspecification; without this, the zero-shot competitiveness claim on real benchmarks risks reducing to benchmark-specific pattern matching rather than robust Bayesian marginalization.

    Authors: We agree that the manuscript does not provide explicit quantitative validation such as graph-property matching between the synthetic prior and the real-world datasets, nor a sensitivity analysis to prior misspecification. The synthetic prior is generated from standard random graph models and causal mechanisms intended to be broadly representative, but this does not substitute for direct validation on PetShop and CausRCA. In the revised manuscript we will add sensitivity analysis (varying graph density, edge probability, and mechanism classes in meta-training) and report resulting performance on the real benchmarks to support the robustness claim. revision: yes

  2. Referee: [§3.2] §3.2 (MACE transformer neural process): the claim of implicitly learning causal structure without model fitting at test time requires explicit confirmation that the attention mechanism over causal structure nodes does not implicitly rely on information leaked from the meta-training distribution that matches the test benchmarks; otherwise the 'parameter-free' zero-shot inference may not hold under distribution shift.

    Authors: The meta-training distribution consists solely of synthetic causal models and is constructed to be disjoint from the structures and data-generating processes in PetShop and CausRCA. The MACE attention operates on causal structure nodes supplied at both training and test time, with parameters frozen after meta-training. To address the referee's concern we will revise §3.2 to state this disjointness explicitly and add controlled synthetic distribution-shift experiments (altering graph properties and mechanisms post-meta-training) that demonstrate maintained zero-shot performance, thereby supporting that the inference does not rely on leaked benchmark-specific information. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The abstract describes PRIM as framing RCA as Bayesian marginalization over a synthetic prior of causal models via simulation-based meta-learning and a MACE transformer neural process, enabling zero-shot inference. No equations, self-citations, or steps are quoted that reduce any reported prediction or identification result to a fitted quantity by construction, nor does the text invoke uniqueness theorems or ansatzes from prior self-work as load-bearing. The representativeness of the synthetic prior is presented as an assumption for generalization to PetShop/CausRCA, not as a definitional equivalence or renamed fit. The derivation chain is therefore self-contained against external benchmarks and the cited prior-fitted networks paradigm.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review yields limited visibility into free parameters or axioms; the central claim rests on the representativeness of the synthetic causal prior and the ability of the meta-training distribution to cover real anomaly patterns.

free parameters (1)
  • meta-training hyperparameters
    Number of synthetic causal models, transformer architecture size, and learning rate schedule are fitted or chosen during meta-training and affect downstream performance.
axioms (1)
  • domain assumption Synthetic causal models generated for meta-training capture the relevant structural uncertainty of target domains.
    Invoked implicitly when claiming generalization from synthetic prior to PetShop and CausRCA without per-domain graph knowledge.

reviewed 2026-06-30 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PRIM: Meta-Learned Bayesian Root Cause Analysis." pith.science (2026). https://pith.science/paper/YDR4FNFU

@misc{pith2026260508786,
  author       = {Pith},
  title        = {Pith review of: PRIM: Meta-Learned Bayesian Root Cause Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YDR4FNFU}},
  note         = {Machine review of arXiv:2605.08786}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Root cause analysis (RCA) in complex systems is challenging due to error propagation across multiple variables, the need for structural causal knowledge, and the computational cost of inference at test time. We introduce PRIM (Prior-fitted Root cause Identification with Meta-learning), a causal meta-learning approach that frames RCA as a Bayesian inference task over a synthetic prior of causal models. By marginalising out structural uncertainty, PRIM implicitly identifies changes in the data-generating mechanism between baseline and anomalous periods. In doing so, PRIM infers distributional differences without explicit statistical testing, and implicitly learns causal structure without model fitting at test time. Following the simulation-based meta-learning paradigm of prior-fitted networks, PRIM uses a Model-Averaged Causal Estimation (MACE) transformer neural process that jointly attends over observational and anomalous samples and the causal structure of nodes, enabling zero-shot inference in 17,ms for systems with up to 100 variables. Across synthetic benchmarks and two realistic benchmark datasets, PetShop and CausRCA, PRIM is competitive with methods that are aware of the system's causal graphical structure a priori while outperforming graph-unaware methods on several tasks. Lightweight fine-tuning to specific domains and data dynamics improves performance further.

Figures

Figures reproduced from arXiv: 2605.08786 by Amadou Ba, Anish Dhir, Bradley Eck, Christopher Lohse, Jonas Wahl, Marco Ruffini.

Figure 1
Figure 1. Figure 1: PRIM architecture. L MACE-TNP blocks refine obs/int embeddings via alternating sample￾and node-level attention. The difference ∆ = H¯ int − H¯ obs is decoded to per-node logits Tˆ ∈ R K. Our model, PRIM (Prior-fitted Root cause Identification with Meta-learning), is built around the MACE-TNP architecture introduced by Dhir et al. [8] for estimating interventional distributions. While the original MACE-TNP … view at source ↗
Figure 2
Figure 2. Figure 2: Three-node confounder vs. mediator scenario. X ( [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Multi-root-cause evaluation on a 6-node DAG (left). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Recall@1 vs. number of nodes at nobs = 100, nint = 10. As highlighted in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the data generation process. A causal graph [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗

Review history (3 revisions) →

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. MATERO-RCA: Mode-Aware Trajectory-Level Energy-Based Root-Set Optimization for Industrial Root Cause Analysis

    eess.SP 2026-07 conditional novelty 6.0

    MATERO-RCA jointly optimizes root set, root-effect modes, and repaired counterfactual trajectories, with a MILP-accelerated best-bound search that reports strong benchmark RCA results.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Bruinsma, and Richard E

    Matthew Ashman, Cristiana Diaconu, Junhyuck Kim, Lakee Sivaraya, Stratis Markou, James Requeima, Wessel P. Bruinsma, and Richard E. Turner. Translation equivariant transformer neural processes. InProceedings of the 41st International Conference on Machine Learning, ICML’24, Vienna, Austria, 2024. JMLR.org

    work page 2024

  2. [2]

    Causal chain analysis and root causes: the giwa approach

    Juan Carlos Belausteguigoitia. Causal chain analysis and root causes: the giwa approach. AMBIO: A Journal of the Human Environment, 33(1):7–12, 2004

    work page 2004

  3. [3]

    Why did the dis- tribution change? InInternational Conference on Artificial Intelligence and Statistics, pages 1666–1674

    Kailash Budhathoki, Dominik Janzing, Patrick Bloebaum, and Hoiyi Ng. Why did the dis- tribution change? InInternational Conference on Artificial Intelligence and Statistics, pages 1666–1674. PMLR, 2021

    work page 2021

  4. [4]

    Causal structure- based root cause analysis of outliers

    Kailash Budhathoki, Lenon Minorics, Patrick Bloebaum, and Dominik Janzing. Causal structure- based root cause analysis of outliers. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba Szepesvari, Gang Niu, and Sivan Sabato, editors,Proceedings of the 39th International Conference on Machine Learning, volume 162 ofProceedings of Machine Learning Resear...

    work page 2022

  5. [5]

    Causeinfer: Automatic and distributed performance diagnosis with hierarchical causality graph in large distributed systems

    Pengfei Chen, Yong Qi, Pengfei Zheng, and Di Hou. Causeinfer: Automatic and distributed performance diagnosis with hierarchical causality graph in large distributed systems. InIEEE INFOCOM 2014-IEEE Conference on Computer Communications, pages 1887–1895. IEEE, 2014

    work page 2014

  6. [6]

    Bcd nets: Scalable variational approaches for bayesian causal discovery.Advances in Neural Information Processing Systems, 34:7095–7110, 2021

    Chris Cundy, Aditya Grover, and Stefano Ermon. Bcd nets: Scalable variational approaches for bayesian causal discovery.Advances in Neural Information Processing Systems, 34:7095–7110, 2021

    work page 2021

  7. [7]

    A meta-learning approach to bayesian causal discovery

    Anish Dhir, Matthew Ashman, James Requeima, and Mark van der Wilk. A meta-learning approach to bayesian causal discovery. In Y . Yue, A. Garg, N. Peng, F. Sha, and R. Yu, editors,International Conference on Learning Representations, volume 2025, pages 14158– 14178, 2025. URL https://proceedings.iclr.cc/paper_files/paper/2025/file/ 24faedc5853648d5857f2cf0...

    work page 2025

  8. [8]

    Turner, and Mark van der Wilk

    Anish Dhir, Cristiana Diaconu, Valentinian Mihai Lungu, James Requeima, Richard E. Turner, and Mark van der Wilk. Estimating interventional distributions with uncertain causal graphs through meta-learning. InThe Thirty-ninth Annual Conference on Neural Information Process- ing Systems, 2025. URLhttps://openreview.net/forum?id=IQlcfc40Ja

    work page 2025

  9. [9]

    Contin- uous bayesian model selection for multivariate causal discovery

    Anish Dhir, Ruby Sedgwick, Avinash Kori, Ben Glocker, and Mark van der Wilk. Contin- uous bayesian model selection for multivariate causal discovery. InForty-second Interna- tional Conference on Machine Learning, 2025. URL https://openreview.net/forum? id=zydNWJzoVd

    work page 2025

  10. [10]

    Interventions and causal inference.Philosophy of science, 74(5):981–995, 2007

    Frederick Eberhardt and Richard Scheines. Interventions and causal inference.Philosophy of science, 74(5):981–995, 2007

    work page 2007

  11. [11]

    Latent bottlenecked attentive neural processes

    Leo Feng, Hossein Hajimirsadeghi, Yoshua Bengio, and Mohamed Osama Ahmed. Latent bottlenecked attentive neural processes. InThe Eleventh International Conference on Learning Representations, 2023. URLhttps://openreview.net/forum?id=yIxtevizEA

    work page 2023

  12. [12]

    Parameter priors for directed acyclic graphical models and the characterization of several probability distributions.The Annals of Statistics, 30(5):1412– 1440, October 2002

    Dan Geiger and David Heckerman. Parameter priors for directed acyclic graphical models and the characterization of several probability distributions.The Annals of Statistics, 30(5):1412– 1440, October 2002. ISSN 0090-5364, 2168-8966. doi: 10.1214/aos/1035844981. URL https: //projecteuclid.org/journals/annals-of-statistics/volume-30/issue-5/ Parameter-prio...

    work page doi:10.1214/aos/1035844981 2002

  13. [13]

    The petshop dataset — finding causes of performance issues across microservices

    Michaela Hardt, William Roy Orchard, Patrick Blöbaum, Elke Kirschbaum, and Shiva Ka- siviswanathan. The petshop dataset — finding causes of performance issues across microservices. In Francesco Locatello and Vanessa Didelez, editors,Proceedings of the Third Conference on 11 Causal Learning and Reasoning, volume 236 ofProceedings of Machine Learning Resear...

    work page 2024

  14. [14]

    Tabpfn: A transformer that solves small tabular classification problems in a second

    Noah Hollmann, Samuel Müller, Katharina Eggensperger, and Frank Hutter. Tabpfn: A transformer that solves small tabular classification problems in a second. InThe Eleventh International Conference on Learning Representations, 2023

    work page 2023

  15. [15]

    Accurate predictions on small data with a tabular foundation model.Nature, 637(8045):319–326, January 2025

    Noah Hollmann, Samuel Müller, Lennart Purucker, Arjun Krishnakumar, Max Körfer, Shi Bin Hoo, Robin Tibor Schirrmeister, and Frank Hutter. Accurate predictions on small data with a tabular foundation model.Nature, 637(8045):319–326, January 2025. ISSN 1476-

    work page 2025

  16. [16]

    Nature , year=

    doi: 10.1038/s41586-024-08328-6. URL https://www.nature.com/articles/ s41586-024-08328-6

    work page doi:10.1038/s41586-024-08328-6

  17. [17]

    From tables to time: Extending TabPFN- v2 to time series forecasting.arXiv preprint arXiv:2501.02945, 2025

    Shi Bin Hoo, Samuel Müller, David Salinas, and Frank Hutter. From tables to time: Extending tabpfn-v2 to time series forecasting.arXiv preprint arXiv:2501.02945, 2025

    work page arXiv 2025

  18. [18]

    Root cause analysis of failures in microservices through causal discovery.Advances in Neural Information Processing Systems, 35:31158–31170, 2022

    Azam Ikram, Sarthak Chakraborty, Subrata Mitra, Shiv Saini, Saurabh Bagchi, and Murat Kocaoglu. Root cause analysis of failures in microservices through causal discovery.Advances in Neural Information Processing Systems, 35:31158–31170, 2022

    work page 2022

  19. [19]

    Root cause analysis of failures from partial causal structures

    Azam Ikram, Kenneth Lee, Shubham Agarwal, Shiv Kumar Saini, Saurabh Bagchi, and Murat Kocaoglu. Root cause analysis of failures from partial causal structures. InThe 41st Conference on Uncertainty in Artificial Intelligence, 2025. URL https://openreview.net/forum?id= 5HeODZrG9E

    work page 2025

  20. [20]

    Causal Discovery from Soft Interventions with Unknown Targets: Characterization and Learn- ing

    Amin Jaber, Murat Kocaoglu, Karthikeyan Shanmugam, and Elias Bareinboim. Causal Discovery from Soft Interventions with Unknown Targets: Characterization and Learn- ing. InAdvances in Neural Information Processing Systems, volume 33, pages 9551–

    work page

  21. [21]

    URL https://papers.nips.cc/paper/2020/hash/ 6cd9313ed34ef58bad3fdd504355e72c-Abstract.html

    Curran Associates, Inc., 2020. URL https://papers.nips.cc/paper/2020/hash/ 6cd9313ed34ef58bad3fdd504355e72c-Abstract.html

    work page 2020

  22. [22]

    A sustainability root cause analysis methodology and its application.Computers & chemical engineering, 35 (12):2786–2798, 2011

    Abhishek Jayswal, Xiang Li, Anand Zanwar, Helen H Lou, and Yinlun Huang. A sustainability root cause analysis methodology and its application.Computers & chemical engineering, 35 (12):2786–2798, 2011

    work page 2011

  23. [23]

    Root cause discovery via permutations and cholesky decomposition.Journal of the Royal Statistical Society Series B: Statistical Methodology, page qkaf066, 2025

    Jinzhou Li, Benjamin B Chu, Ines F Scheller, Julien Gagneur, and Marloes H Maathuis. Root cause discovery via permutations and cholesky decomposition.Journal of the Royal Statistical Society Series B: Statistical Methodology, page qkaf066, 2025

    work page 2025

  24. [24]

    Nassar, and Peter W

    Mingjie Li, Zeyan Li, Kanglin Yin, Xiaohui Nie, Wenchi Zhang, Kaixin Sui, and Dan Pei. Causal inference-based root cause analysis for online service systems with intervention recognition. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, page 3230–3240, New York, NY , USA, 2022. Association for Computing Mac...

    work page doi:10.1145/3534678.3539041 2022

  25. [25]

    MetaRCA: A Generalizable Root Cause Analysis Framework for Cloud-Native Systems Powered by Meta Causal Knowledge, March 2026

    Shuai Liang, Pengfei Chen, Bozhe Tian, Gou Tan, Maohong Xu, Youjun Qu, Yahui Zhao, Yiduo Shang, and Chongkang Tan. MetaRCA: A Generalizable Root Cause Analysis Framework for Cloud-Native Systems Powered by Meta Causal Knowledge, March 2026. URL http: //arxiv.org/abs/2603.02032. arXiv:2603.02032 [cs]

    work page arXiv 2026

  26. [26]

    Microscope: Pinpoint performance issues with causal graphs in micro-service environments

    JinJin Lin, Pengfei Chen, and Zibin Zheng. Microscope: Pinpoint performance issues with causal graphs in micro-service environments. InInternational Conference on Service-Oriented Computing, pages 3–20. Springer, 2018

    work page 2018

  27. [27]

    Microhecl: High-efficient root cause localization in large-scale microservice systems

    Dewei Liu, Chuan He, Xin Peng, Fan Lin, Chenxi Zhang, Shengfang Gong, Ziang Li, Jiayu Ou, and Zheshun Wu. Microhecl: High-efficient root cause localization in large-scale microservice systems. In2021 IEEE/ACM 43rd International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP), pages 338–347. IEEE, 2021

    work page 2021

  28. [28]

    Licence to scale: A microservice simulation environment for benchmarking agentic ai

    Christopher Lohse, Adrian Selk, Amadou Ba, Jonas Wahl, and Marco Ruffini. Licence to scale: A microservice simulation environment for benchmarking agentic ai. Accepted at NeurIPS 2025 Workshop Scaling Environments for Agents (SEA), December 2025. 12

    work page 2025

  29. [29]

    Dibs: Differentiable bayesian structure learning.Advances in Neural Information Processing Systems, 34:24111– 24123, 2021

    Lars Lorch, Jonas Rothfuss, Bernhard Schölkopf, and Andreas Krause. Dibs: Differentiable bayesian structure learning.Advances in Neural Information Processing Systems, 34:24111– 24123, 2021

    work page 2021

  30. [30]

    Patterns of unexpected in-hospital deaths: a root cause analysis.Patient safety in surgery, 5(1):3, 2011

    Lawrence A Lynn and J Paul Curry. Patterns of unexpected in-hospital deaths: a root cause analysis.Patient safety in surgery, 5(1):3, 2011

    work page 2011

  31. [31]

    Enabling Joint Benchmarking of Automated Root Cause Analysis and Causal Discovery in Manufac- turing Using thecausRCADataset.Procedia CIRP, 139:114–120, January 2026

    Carl Willy Mehling, Sven Pieper, Tobias Lüke, Julius Döbelt, and Steffen Ihlenfeldt. Enabling Joint Benchmarking of Automated Root Cause Analysis and Causal Discovery in Manufac- turing Using thecausRCADataset.Procedia CIRP, 139:114–120, January 2026. ISSN 2212-8271. doi: 10.1016/j.procir.2025.09.010. URL https://www.sciencedirect.com/ science/article/pii...

    work page doi:10.1016/j.procir.2025.09.010 2026

  32. [32]

    Transformers can do bayesian inference

    Samuel Müller, Noah Hollmann, Sebastian Pineda Arango, Josif Grabocka, and Frank Hutter. Transformers can do bayesian inference. InInternational Conference on Learning Representa- tions, 2022. URLhttps://openreview.net/forum?id=KSugKcbNf9

    work page 2022

  33. [33]

    Transformer neural processes: Uncertainty-aware meta learning via sequence modeling.ArXiv, abs/2207.04179, 2022

    Tung Nguyen and Aditya Grover. Transformer neural processes: Uncertainty-aware meta learning via sequence modeling.ArXiv, abs/2207.04179, 2022. URL https://api. semanticscholar.org/CorpusID:250340974

    work page arXiv 2022

  34. [34]

    Root cause analysis of outliers with missing structural knowledge

    William Roy Orchard, Nastaran Okati, Sergio Hernan Garrido Mejia, Patrick Blöbaum, and Dominik Janzing. Root cause analysis of outliers with missing structural knowledge. In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2026. URL https://openreview.net/forum?id=7Nxq4RQApu

    work page 2026

  35. [35]

    The pagerank citation ranking: Bringing order to the web

    Lawrence Page, Sergey Brin, Rajeev Motwani, and Terry Winograd. The pagerank citation ranking: Bringing order to the web. Technical report, Stanford infolab, 1999

    work page 1999

  36. [36]

    Causal diagrams for empirical research.Biometrika, 82(4):669–688, 1995

    Judea Pearl. Causal diagrams for empirical research.Biometrika, 82(4):669–688, 1995

    work page 1995

  37. [37]

    Cambridge university press, 2009

    Judea Pearl.Causality. Cambridge university press, 2009

    work page 2009

  38. [38]

    Luan Pham, Huong Ha, and Hongyu Zhang. Baro: Robust root cause analysis for microservices via multivariate bayesian online change point detection.Proceedings of the ACM on Software Engineering, 1(FSE):2214–2237, July 2024. ISSN 2994-970X. doi: 10.1145/3660805

    work page doi:10.1145/3660805 2024

  39. [39]

    Do-PFN: In-context learning for causal effect estimation

    Jake Robertson, Arik Reuter, Siyuan Guo, Noah Hollmann, Frank Hutter, and Bernhard Schölkopf. Do-PFN: In-context learning for causal effect estimation. InThe Thirty- ninth Annual Conference on Neural Information Processing Systems, 2026. URL https: //openreview.net/forum?id=OaNbl9b56B

    work page 2026

  40. [40]

    Root cause analysis of outliers in unknown cyclic graphs.arXiv preprint arXiv:2510.06995, 2025

    Daniela Schkoda and Dominik Janzing. Root cause analysis of outliers in unknown cyclic graphs.arXiv preprint arXiv:2510.06995, 2025

    work page internal anchor Pith review arXiv 2025

  41. [41]

    Impact of gene mutation in the develop- ment of parkinson’s disease.Genes & diseases, 6(2):120–128, 2019

    Suganya Selvaraj and Shanmughavel Piramanayagam. Impact of gene mutation in the develop- ment of parkinson’s disease.Genes & diseases, 6(2):120–128, 2019

    work page 2019

  42. [42]

    ϵ-diagnosis: Unsupervised and real-time diagnosis of small-window long-tail latency in large-scale microservice platforms

    Huasong Shan, Yuan Chen, Haifeng Liu, Yunpeng Zhang, Xiao Xiao, Xiaofeng He, Min Li, and Wei Ding. ϵ-diagnosis: Unsupervised and real-time diagnosis of small-window long-tail latency in large-scale microservice platforms. InThe World Wide Web Conference, WWW ’19, page 3215–3222, New York, NY , USA, 2019. Association for Computing Machinery. ISBN 978145036...

    work page doi:10.1145/3308558.3313653 2019

  43. [43]

    Anomaly detection and failure root cause analysis in (micro) service-based cloud applications: A survey.ACM Comput

    Jacopo Soldani and Antonio Brogi. Anomaly detection and failure root cause analysis in (micro) service-based cloud applications: A survey.ACM Comput. Surv., 55(3), February 2022. ISSN 0360-0300. doi: 10.1145/3501297. URLhttps://doi.org/10.1145/3501297

    work page doi:10.1145/3501297 2022

  44. [44]

    Causal discovery and inference: concepts and recent methodolog- ical advances

    Peter Spirtes and Kun Zhang. Causal discovery and inference: concepts and recent methodolog- ical advances. InApplied informatics, volume 3, pages 1–28. Springer, 2016

    work page 2016

  45. [45]

    MIT press, 2000

    Peter Spirtes, Clark N Glymour, and Richard Scheines.Causation, prediction, and search. MIT press, 2000. 13

    work page 2000

  46. [46]

    Permutation-based causal structure learning with unknown intervention targets

    Chandler Squires, Yuhao Wang, and Caroline Uhler. Permutation-based causal structure learning with unknown intervention targets. InConference on Uncertainty in Artificial Intelligence, pages 1039–1048. PMLR, 2020

    work page 2020

  47. [47]

    Scalable intervention target estimation in linear models

    Burak Varici, Karthikeyan Shanmugam, Prasanna Sattigeri, and Ali Tajer. Scalable intervention target estimation in linear models. In M. Ranzato, A. Beygelzimer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, volume 34, pages 1494–1505. Curran Associates, Inc., 2021. URLhttps://proceedings.neurip...

    work page 2021

  48. [48]

    Intervention target estimation in the presence of latent variables

    Burak Varici, Karthikeyan Shanmugam, Prasanna Sattigeri, and Ali Tajer. Intervention target estimation in the presence of latent variables. In James Cussens and Kun Zhang, editors, Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, volume 180 ofProceedings of Machine Learning Research, pages 2013–2023. PMLR, 01–05 Aug 2...

    work page 2013

  49. [49]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    work page 2017

  50. [50]

    Characterizing and learning equivalence classes of causal dags under interventions

    Karren Yang, Abigail Katcoff, and Caroline Uhler. Characterizing and learning equivalence classes of causal dags under interventions. InInternational Conference on Machine Learning, pages 5541–5550. PMLR, 2018

    work page 2018

  51. [51]

    Learning unknown intervention targets in structural causal models from heterogeneous data

    Yuqin Yang, Saber Salehkaleybar, and Negar Kiyavash. Learning unknown intervention targets in structural causal models from heterogeneous data. In Sanjoy Dasgupta, Stephan Mandt, and Yingzhen Li, editors,Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, volume 238 ofProceedings of Machine Learning Research, pages ...

    work page 2024

  52. [52]

    Sample functional mechanisms

    work page

  53. [53]

    Sample observational data Sample in topological order:

    work page

  54. [54]

    Hard do-intervention (stuck sensor) Pin node to a constant:

    Select intervention node(s) Prefer non-leaf nodes (nodes with at least one causal successor) OR 5b. Hard do-intervention (stuck sensor) Pin node to a constant:

    work page

  55. [55]

    stuck sensor

    Sample interventional data Sample from modified SCM in topological order 5a. Soft intervention (weight, function, sign or noisechange) Modify SCM Figure 5: Overview of the data generation process. A causal graph G and functional mechanisms f are sampled to define an SCM. Observational data Dobs is drawn by ancestral sampling, after which a target node tj ...

    work page

This paper was first reviewed by grok-4.3 on June 30, 2026.