Natural Counterfactuals With Necessary Backtracking

Authors: GUANG-YUAN HAO, Jiji Zhang, Biwei Huang, Hao Wang, Kun Zhang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the effectiveness of our method through empirical experiments on four synthetic datasets and two real-world datasets.
Researcher Affiliation Academia 1Chinese University of Hong Kong, 2University of California San Diego, 3Rutgers University 4Carnegie Mellon University, 5Mohamed bin Zayed University of Artificial Intelligence
Pseudocode No The paper describes the proposed method but does not include a formal pseudocode or algorithm block.
Open Source Code Yes The code is available at https: //github.com/Guangyuan Hao/natural_counterfactuals.
Open Datasets Yes We design four simulation datasets, Toy 1-4, and use the designed SCMs to generate 10, 000 data points as a training dataset and another 10, 000 data points as a test set for each dataset.
Dataset Splits No The paper specifies training and test sets but does not explicitly mention or detail a validation set or its split for the experiments.
Hardware Specification Yes All the experiments above were run on NVIDIA RTX 4090 GPUs.
Software Dependencies No The paper mentions software components like 'normalizing flows', 'Adam W optimizer', 'V-SCM', and 'H-SCM', but does not provide specific version numbers for these or other key software dependencies.
Experiment Setup Yes Our training regimen for the flow-based model spanned 2000 epochs, utilizing a batch size of 100 in conjunction with the Adam W optimizer. We initialized the learning rate to 10 3, set β1 to 0.9, β2 to 0.9.