TO COMPLETE. I kinda dropped this one cause it’s really academic and isn’t a common type of competition

Link: https://www.kaggle.com/c/neurips-2023-machine-unlearning

Problem Type:

Input: model weights

Output: model weights that have unlearned images from the “forget set”

Eval Metric:

  • To measure forgetting quality we require an estimation of two distributions: the one obtained from unlearning and that obtained from retraining-from-scratch.

Summary

Important notebooks/discussions:

Takeaways