Multi-domain Causal Structure Learning in Linear Systems

Authors: AmirEmad Ghassami, Negar Kiyavash, Biwei Huang, Kun Zhang

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The performance of the proposed methods are depicted in Figure 1. We have depicted the F1 score for each case. As seen from the F1 score, the IB method performs better than the MC method in the first model. However, if in an application we know that the parameters are not likely to change much (as in model 2), the MC method also has high performance. Moreover, for the case of limited number of domains, the IB method significantly outperforms using HSIC test. We also tested our proposed methods when a latent confounder was present in the system. As could be seen in Figure 1, IB is generally more robust to the presence of latent confounder.
Researcher Affiliation Academia Department of ECE, University of Illinois at Urbana-Champaign, Urbana, IL, USA. School of ISy E and ECE, Georgia Institute of Technology, Atlanta, GA, USA. Department of Philosophy, Carnegie Mellon University, Pittsburgh, PA, USA.
Pseudocode Yes Algorithm 1 Multi-domain Causal Structure Learning. ... Algorithm 2 MC Causal Structure Learning.
Open Source Code No The paper does not contain any explicit statements about releasing source code for the methodology, nor does it provide a link to a code repository.
Open Datasets Yes We applied our methods to f MRI hippocampus dataset [PL], which contains signals from six separate brain regions: perirhinal cortex (PRC), parahippocampal cortex (PHC), entorhinal cortex (ERC), subiculum (Sub), CA1, and CA3/Dentate Gyrus (CA3) in the resting states on the same person in 84 successive days. We used the anatomical connections [BB08, ZHZ+17] as a reference. ... [PL] Poldrack and Laumann. https://openfmri.org/dataset/ds000031/, 2015.
Dataset Splits No The paper mentions the number of samples per domain (103) and thresholds used for graph construction, but it does not specify explicit train/validation/test splits for the datasets used in evaluation.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions using the 'state-of-the-art HISC test [GFT+08]' and comparing with 'Li NGAM Algorithm [SHHK06]', which are methods. However, it does not specify any software names with version numbers required for reproducibility (e.g., Python 3.x, PyTorch 1.x, specific library versions).
Experiment Setup Yes The number of samples in each domain is 103. We have used the state-of-the-art HISC test [GFT+08] as our non-parametric independence test. ... We set a threshold = 0.1 on B from each domain; if |Bij| is larger than , then there is an edge from Xi to Xj. Then if an edge appears in more than 80% of all domains, we take this edge in the final graph. ... if an edge appears in more than 60% of all domains, we took this edge in the final graph.