Offline Generative Active Learning: Feasibility and Limitations

Offline experiments confirm the feasibility of gradient-based contribution estimation by analyzing noise effects on CIFAR-10.