Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup
Authors: Damien Teney, Jindong Wang, Ehsan Abbasnejad
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4. Empirical Verification We performed a large number of experiments to understand the contribution of the different effects of selective mixup and other resampling baselines (complete results in Appendix C). |
| Researcher Affiliation | Collaboration | 1Idiap Research Institute, Martigny, Switzerland 2Microsoft Research Asia, Beijing, China 3AIML, University of Adelaide, Australia. |
| Pseudocode | No | No explicitly labeled pseudocode or algorithm blocks were found. |
| Open Source Code | No | No explicit statement of code release or a direct link to a code repository for the methodology was found. |
| Open Datasets | Yes | Waterbirds (Sagawa et al., 2019) is a popular artificial dataset used to study distribution shifts. |
| Dataset Splits | Yes | All experiments use early stopping i.e. recording metrics for each run at the epoch of highest ID or worst-group validation performance (for Wild-Time and waterbirds/civil Comments datasets respectively). |
| Hardware Specification | Yes | All experiments were run on a single laptop with an Nvidia GeForce RTX 3050 Ti GPU. |
| Software Dependencies | No | The paper mentions "standard architectures" and "pretrained BERT" and implies the use of frameworks like PyTorch (given the GPU mention), but it does not specify version numbers for any software dependencies. |
| Experiment Setup | Yes | We follow prior work on each dataset for the architectures and hyperparameters of our experiments. |