Multi-Granularity Causal Structure Learning
Authors: Jiaxuan Liang, Jun Wang, Guoxian Yu, Shuyin Xia, Guoyin Wang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that Mg CSL outperforms competitive baselines, and finds out explainable causal connections on f MRI datasets. |
| Researcher Affiliation | Academia | Jiaxuan Liang1, 2, Jun Wang2, Guoxian Yu1, 2, *, Shuyin Xia3, 4, Guoyin Wang3, 4 1School of Software, Shandong University, Jinan, China 2SDU-NTU Joint Centre for AI Research, Shandong University, Jinan, China 3Chongqing Key Laboratory of Computational Intelligence, Chongqing Uni. of Posts and Telecom., Chongqing, China 4MOE Key Laboratory of Big Data Intelligent Computing, Chongqing Uni. of Posts and Telecom., Chongqing, China |
| Pseudocode | Yes | The comprehensive procedure of Mg CSL is delineated in Algorithm 1 in the Supplementary file. |
| Open Source Code | Yes | The code of Mg CSL is shared at http://www.sdu-idea.cn/codes.php?name=Mg CSL. |
| Open Datasets | Yes | We consider another well-known dataset Sachs (Sachs et al. 2005), which measures the level of different expressions of proteins and phospholipids in human cells. The ground truth causal graph of this dataset consists of 11 variables and 20 edges. In this work, we test on observational data with 7466 samples. |
| Dataset Splits | No | The paper mentions generating datasets of a certain sample size (e.g., '10 datasets of n=1000 samples') and then evaluating the estimated DAG. It does not explicitly specify how these datasets are split into distinct training, validation, and test subsets for model learning and evaluation within the experimental setup. |
| Hardware Specification | No | The paper does not explicitly specify any hardware used for running the experiments (e.g., GPU models, CPU types, or memory details). |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies or libraries used in the implementation or experiments. |
| Experiment Setup | Yes | We carry out parameter sensitivity study w.r.t. α1 and α2. A small α1 may introduce considerable noise, while an excessively large α1 could weaken the representational capacity of macro-variables. Besides, a larger α2 can improve the performance in terms of precision and running time, but Mg CSL tends to achieve an overly sparse estimated graph with a few edges when α2 is too large. Based on the aforementioned analysis, we set α1=0.1 and α2=0.01. |