1 paper
Andrada Gobeaja, Ionut Hodoroaga, Elena Burceanu +1
Can a dataset be recognized from the spurious correlations it induces during training? We argue that datasets leave dataset-specific traces in a model's learned semantic correlatio…