An energy-reweighted flow-matching loss is shown by construction to target a reweighted distribution, with modest benchmark demonstrations of concept removal.
Generation from the full distribution (left), and after suppressing all classes except automobile (middle) and airplane (right), which are assigned low energy
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
-
ContinualFlow: Learning and Unlearning with Neural Flow Matching
An energy-reweighted flow-matching loss is shown by construction to target a reweighted distribution, with modest benchmark demonstrations of concept removal.