TAD combines a loss-based trusted/untrusted split with anchor-based pseudo-label recalibration to improve dataset distillation accuracy under synthetic and real label noise.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Trust-Aware Diversion for Data-Effective Distillation
TAD combines a loss-based trusted/untrusted split with anchor-based pseudo-label recalibration to improve dataset distillation accuracy under synthetic and real label noise.