iSCAN: Identifying Causal Mechanism Shifts among Nonlinear Additive Noise Models

Authors: Tianyu Chen, Kevin Bello, Bryon Aragam, Pradeep Ravikumar

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and real-world data are provided to showcase the applicability of this approach.
Researcher Affiliation Academia Department of Statistics and Data Science, University of Texas at Austin Booth School of Business, University of Chicago Machine Learning Department, Carnegie Mellon University
Pseudocode Yes Algorithm 1 i SCAN Identifying Shifts in Causal Additive Noise models. Input: Datasets X1, . . . , XH. Output: Shifted variables set b S, and topological sort ˆπ.
Open Source Code Yes Code implementing the proposed method is open-source and publicly available at https://github.com/kevinsbello/i SCAN.
Open Datasets No For the synthetic experiments, data generation process is described, but no public dataset or repository link is provided for the generated data. For the apoptosis data, it mentions using “an ovarian cancer dataset” previously analyzed by others but does not provide a direct link or citation for accessing it.
Dataset Splits No The paper describes data generation for synthetic experiments (e.g., “500 data points per environment”) but does not specify any train/validation/test splits or cross-validation setup for the datasets used.
Hardware Specification Yes The running time was recorded by executing the experiments on an Intel Xeon Gold 6248R Processor with 8 cores.
Software Dependencies Yes To automatically select the elbow we made use of the Python package Kneed7, with hyperparameters curve= convex , direction= decreasing , online=online, interp_method= interp1d . (Footnote 7: We used the latest version found at: https://kneed.readthedocs.io/en/stable/)
Experiment Setup Yes For our method, we used η = 0.05 for eq.(6) and eq.(7), and a threshold t = 2 (see Alg. 3).