Learning Causal Semantic Representation for Out-of-Distribution Prediction

Authors: Chang Liu, Xinwei Sun, Jindong Wang, Haoyue Tang, Tao Li, Tao Qin, Wei Chen, Tie-Yan Liu

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical study shows improved OOD performance over prevailing baselines. We develop effective methods for OOD generalization and domain adaptation, and achieve mostly better performance than prevailing methods on real-world image classification tasks.
Researcher Affiliation Collaboration 1 Microsoft Research Asia, Beijing, 100080. 2 Tsinghua University, Beijing, 100084. 3 Peking University, Beijing, 100871.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks. Methods are described textually and mathematically.
Open Source Code Yes Codes are available at https://github.com/changliu00/causal-semantic-generative-model.
Open Datasets Yes Shifted-MNIST. Image CLEF-DA is a standard benchmark for domain adaptation [1]. PACS is a more recent benchmark dataset [69]. VLCS [30].
Dataset Splits Yes For our methods, we use a validation set from the training domain for model selection.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. It only discusses model architectures and datasets.
Software Dependencies No The paper mentions 'PyTorch' [84] but does not specify a version number for it or any other ancillary software components, which is necessary for reproducibility.
Experiment Setup Yes In practice x often has a much larger dimension than y, making the first supervision term overwhelmed by the second unsupervised term in Eqs. (2,3,5). So we downscale the second term.