Please verify the version of the WILDS package that you used.

Please provide a contact name that we will use to list on the leaderboard.

The email address that people should contact you at if they have questions. This will be made public on the leaderboard.

What institution/organization you are from, if you'd like to provide this information.

Is your implementation official (i.e., you proposed this method, or you made substantial modifications to an existing method and have written that up), or unofficial (i.e., you re-implemented a method that others had proposed)?

The name of the algorithm you used (e.g., "ERM", "Group DRO").

Which validation sets did you use? If "other", please describe how you did model/hyperparameter selection.

Please provide a publicly-accessible URL to the .tar.gz or .zip file containing your submission.

Please link to the original paper that describes your method. Please be sure to only include the URL.

Please link to a Github or other public repository that contains all of the code and scripts required to reproduce your results. Do not put a placeholder.

Indicate if the submission is standard or non-standard. Please read the rules concerning standard vs. non-standard submissions at http://wilds.stanford.edu/submit.

 If the submission is non-standard, please describe what changes you made from the standard protocol.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "BERT-base-uncased").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "BERT-base-uncased").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "Faster R-CNN").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

If you used any unlabeled data for training your model or for doing hyperparameter search, please indicate so below. If you've used other sources of unlabeled data that aren't from the official WILDS dataset (e.g., from other existing datasets or scraped from the Internet) and that are not part of the official WILDS default model pretraining, please describe them briefly.

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.

If you used the default model in the WILDS package, select "Default". Otherwise, provide the name of the model you used (e.g., "ResNet50").

Please disclose the hyperparameters you selected and how you tuned them. If you did a grid search, please follow this format: "lr: [0.001*, 0.01*], dropout: [0*, 0.5], etc.", where the asterisks denote the hyperparameters that you eventually selected based on validation performance. If you did a random search, please let us know what ranges you searched over and how many hyperparameter combinations you tried. This information is not currently displayed on the leaderboards, but it is important for record-keeping.