Research map / Probabilistic and scientific modeling / Causal discovery

Root Cause Analysis of Measurement and Mechanistic Anomalies

NeurIPS 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Causal effect identification · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Formalizes measurement vs mechanistic anomalies as latent/observed interventions in a causal DAG and proposes CALI, a hard-assignment likelihood method for root cause localization and anomaly-type classification.

Problem

Root cause analysis of anomalies usually identifies which features are responsible but ignores whether an anomaly is a measurement error or a change in the underlying causal mechanism.

Approach

Model measurement anomalies as interventions on observed readouts and mechanistic anomalies as interventions on latent causal states, infer per-sample assignments via a sparse (Bernoulli-prior) hard-assignment likelihood, and score root causes by likelihood gain and type by likelihood difference.

Key results

• On synthetic data, CALI* with true graph achieves the best root-cause Top-k recall, with CALI on a learned graph a close second.
• CALI* matches exhaustive model-optimal classifier Opt* in localization and type accuracy while running 47.57s vs 1190.60s on average (500 samples, 3 nodes).
• On Causal Chambers (medium outliers), CALI* reaches 90% Top-k recall; CALI beats the marginal baseline Marg* by a large margin in type classification under weak outliers on Sachs.

Surprising findings

• Marginal-outlierness baselines can perform competitively under extremely large anomalous shifts, where density estimates extrapolate poorly.

Limitations

• Requires a DAG; performance depends on causal discovery quality, which contamination may degrade.
• Degrades under severe model misspecification; density estimates unreliable for extremely large shifts.
• Restricted to tabular data.

Experimental setup

Datasets: Sachs Causal Chambers NYC Yellow Taxi Trip Dataset

Benchmarks: Sachs Causal Chambers synthetic ER-DAG SCM NYC Yellow Taxi (case study)

Metrics: Top-k recall anomaly-type classification accuracy

Baselines: SCM attribution (SCM*) SmoothTraversal / ScoreOrder (SmTr*, ScOr) AE-Shap Marg* Opt*

Builds on: Sparse Mechanism Shift Independence of Causal Mechanisms additive noise model NoGAM Shapley-value SCM attribution

Closest papers by meaning

• Root Cause Analysis of Outliers with Missing Structural Knowledge (NeurIPS 2025 · similarity 0.92)
• Robust Root Cause Diagnosis using In-Distribution Interventions (ICLR 2025 · similarity 0.86)
• Where Root Cause Analysis Fails: A Retrieval-Reranking Decomposition (NeurIPS 2026 · similarity 0.84)
• PRIM: Meta-Learned Bayesian Root Cause Analysis (NeurIPS 2026 · similarity 0.83)
• Detecting and Measuring Confounding Using Causal Mechanism Shifts (NeurIPS 2024 · similarity 0.81)
• Local Identifying Causal Relations in the Presence of Latent Variables (ICML 2025 · similarity 0.81)