Root Cause Explanation of Outliers under Noisy Mechanisms
Authors: Phuoc Nguyen, Truyen Tran, Sunil Gupta, Thin Nguyen, Svetha Venkatesh
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on simulated datasets and two real-world scenario datasets show better anomaly attribution performance of the proposed method compared to the baselines. Our method scales to larger graphs with more nodes and edges. |
| Researcher Affiliation | Academia | A2I2, Deakin University {phuoc.nguyen,truyen.tran,sunil.gupta,thin.nguyen,svetha.venkatesh}@deakin.edu.au |
| Pseudocode | No | No pseudocode or algorithm blocks are provided. |
| Open Source Code | No | No statement explicitly providing concrete access to source code for the described methodology. |
| Open Datasets | No | The paper mentions using 'simulated datasets and two real-world scenario datasets' (micro cloud service and supply chain) but does not provide specific links, DOIs, repositories, or formal citations for public access to these datasets. Simulated data is generated, not publicly accessed. |
| Dataset Splits | No | The paper mentions 'training data collected during the normal operation' but does not specify exact train/validation/test splits (e.g., percentages, sample counts, or explicit cross-validation setup). |
| Hardware Specification | Yes | Fig. 2 shows the runtime of all methods on a Ubuntu 20.04 workstation with an Intel Xeon E5-1650 v4 CPU and 46Gb RAM. |
| Software Dependencies | No | The paper mentions 'Ubuntu 20.04 workstation' but does not specify other software dependencies with version numbers (e.g., specific libraries, frameworks, or programming language versions). |
| Experiment Setup | Yes | The noisy causal mechanisms follow Definition 1 with α 1 j = 1, β 1 ij = 0.01, and µij |N(0, 1)|. We draw normal data from these noisy FCMs, following their generative process. For the abnormal data, we randomly select a target node Xn from this causal graph. We then choose among its ancestor node either k [1, . . . , m] root-cause nodes, or l [1, . . . , m] root-cause edges, or k + l root-cause nodes and edges. Here, m is chosen to be 10% of the number of nodes in the subgraph. We inject outlier noises into the nodes and edges to create the ground truths as follows: The outlier node noise ϵj of node j is randomly drawn from N(a, b), where a is drawn from Uniform(3, 5) and b is drawn from Uniform(3, 5). The outlier edge noise ξij of each node j is randomly drawn from N(amj, bsj), where a is drawn from Uniform(3, 5), b is drawn from Uniform(3, 5), and mj represents the maximum magnitude of the current weights wij (i.e., mj = maxi(|wij|)), and sj represents the maximum standard deviation σij (i.e., sj = maxi(σij)). |