The paper proposes a trustworthy DFML framework that mitigates task-distribution shift and corruption via synthetic task reconstruction, historical task replay through interpolation, and automatic filtering of untrustworthy models.
Up to 100x faster data-free knowledge distillation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2023 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Task-Distributionally Robust Data-Free Meta-Learning
The paper proposes a trustworthy DFML framework that mitigates task-distribution shift and corruption via synthetic task reconstruction, historical task replay through interpolation, and automatic filtering of untrustworthy models.