Causal Abstraction Inference under Lossy Representations
Authors: Kevin Muyuan Xia, Elias Bareinboim
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we experimentally show the effectiveness of projected abstraction models in high-dimensional image settings. ... In Sec. 4, we empirically demonstrate the power of abstractions at performing causal inference in high-dimensional image settings. ... We perform two experiments to demonstrate the benefits of projected abstractions. The models in the experiments leverage Neural Causal Models (NCMs) (Xia et al., 2021; 2023), specifically the generative adversarial implementation called GAN-NCMs. ... The results are shown in Fig. 6. |
| Researcher Affiliation | Academia | 1Causal AI Lab, Columbia University. Correspondence to: Kevin Xia <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Constructing MH from ML. |
| Open Source Code | Yes | Details of the experiment setup can be found in App. D, and code can be found at https://github.com/Causal AILab/ Projected Causal Abstractions. |
| Open Datasets | Yes | The high-level query τ(Q) = P(yx | z) is estimated in the graph setting shown in Fig. 3(a), where Z is a digit from 0 to 9, X is a corresponding colored MNIST image, and Y is a label denoting the color prediction of X. ... Figure 6: Mean absolute error (MAE) v. number of samples for the MNIST estimation task. |
| Dataset Splits | No | The paper mentions using the MNIST dataset, a well-known benchmark, but does not explicitly state the specific train/test/validation split percentages, sample counts, or detailed methodology used for splitting in the main text. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers, such as programming language versions or library versions (e.g., Python 3.8, PyTorch 1.9). |
| Experiment Setup | No | The paper states: "Details of the experiment setup can be found in App. D", but does not provide specific experimental setup details, such as concrete hyperparameter values or training configurations, in the main text. |