Invariant Risk Minimization Is A Total Variation Model
Authors: Zhao-Rong Lai, Weiwen Wang
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that the proposed framework achieves competitive performance in several benchmark machine learning scenarios. (Abstract, Section 4. Experiments, Section 4.1. Simulation Study, Section 4.2. Real-world Experiments) |
| Researcher Affiliation | Academia | 1Department of Mathematics, College of Information Science and Technology, Jinan University, Guangzhou, China. Correspondence to: Weiwen Wang <wangww29@jnu.edu.cn>. |
| Pseudocode | No | The paper describes update formulas for optimization (e.g., 'Φ(k+1) = Φ(k) η Φg(Φ(k))') in Appendix B.1 but does not present them in a clearly labeled 'Pseudocode' or 'Algorithm' block with structured steps. |
| Open Source Code | Yes | Code is available at https://github.com/laizhr/IRM-TV. |
| Open Datasets | Yes | We use the House Prices data set1 to verify the TV-ℓ1-based models in a regression task... 1https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data (Section 4.2) This data set contains face images of celebrities (Liu et al., 2015). (Section 4.2) The Landcover data set records time series and the corresponding land cover types from the satellite data (Gislason et al., 2006; Russwurm et al., 2020; Xie et al., 2021). (Section 4.2) In this task we use the Adult data set2 to predict if the income of an individual exceeds $50K/yr based on the census data... 2https://archive.ics.uci.edu/dataset/2/adult (Section 4.2) |
| Dataset Splits | Yes | In training sample generation, ps(t) is fixed as p s for t [0, 0.5) and as p+ s for t [0.5, 1]... (Section 4.1) Samples with built year in period [1900, 1950] are used for training and those with built year in period (1950, 2000] are used for test. (Section 4.2, House Price Prediction) We randomly choose two thirds of data from the subgroups Black Male and Non-Black Female for training, and then verify models across all four subgroups with the rest data. (Section 4.2, Adult Income Prediction) |
| Hardware Specification | No | The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Pytorch' (Appendix B.1) but does not specify its version number or any other software dependencies with their specific versions. |
| Experiment Setup | Yes | More implementing details can be found in the code link, such as the learning rate, the number of training epochs, etc. (Appendix B.1) We apply min-batch subgradients with batch size 1024 in Landcover, and full-batch subgradients in the other data sets. (Appendix B.1) Table B2 provides Pytorch-style architectures of the invariant feature extractor Φ and the environment inferring measure ρ, detailing specific layers and activation functions for each dataset. (Appendix B.3) |