Contact: Mahsa Forouzesh
For time-series applications, cross-validation is not valid because random-shuffling over the data gives imbalanced subsets that might be dependent on each other. In this project, we would like to explore possible replacements for cross-validation in such settings. In particular, we would like to study the potential benefits of gradient disparity (introduced in ) as an early stopping criterion that does not require a held-out validation set.
 Forouzesh and Thiran, Disparity Between Batches as a Signal for Early Stopping, ECML/PKDD 2021
To apply please send your CV and transcript to Mahsa Forouzesh