A new data cartography framework that uses early-epoch loss and forget-event frequency to identify memorization hotspots and guide pruning in generative model training.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Data Cartography for Detecting Memorization Hotspots and Guiding Data Interventions in Generative Models
A new data cartography framework that uses early-epoch loss and forget-event frequency to identify memorization hotspots and guide pruning in generative model training.