Can Subnetwork Structure Be the Key to Out-of-Distribution Generalization?

Authors: Dinghuai Zhang, Kartik Ahuja, Yilun Xu, Yisen Wang, Aaron Courville

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on various OOD generalization tasks corroborate the effectiveness of our method.
Researcher Affiliation Academia 1Mila Quebec AI Institute 2CSAIL, Massachusetts Institute of Technology 3Key Lab of Machine Perception (Mo E), School of EECS, Peking University.
Pseudocode Yes Algorithm 1 Modular Risk Minimization
Open Source Code No The paper states 'We defer all other experimental details to the supplementary materials' but does not explicitly state that code is open-source or provide a specific link.
Open Datasets Yes We take the intuition from Arjovsky et al. (2019); Nam et al. (2020); Ahuja et al. (2021); Ahmed et al. (2021) to design a biased variant of the original MNIST dataset (Le Cun et al., 1998).
Dataset Splits No The paper defines 'seen environments' for training and 'unseen environments' for testing, and splits the 'out-domain' into 'in-split' and 'out-split'. However, it does not provide specific numerical details (percentages or counts) for standard training, validation, and test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment.
Experiment Setup No The paper mentions 'For all methods we keep the same model architectures and training settings. We defer all other experimental details to the supplementary materials.' This indicates details exist but are not provided explicitly in the main text.