A replay-based diffusion training scheme with three consistency losses reduces generative forgetting on several 32x32 image benchmarks, though the supporting upper bound is not rigorously established.
Efficient lifelong learning with a-gem
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CCD: Continual Consistency Diffusion for Lifelong Generative Modeling
A replay-based diffusion training scheme with three consistency losses reduces generative forgetting on several 32x32 image benchmarks, though the supporting upper bound is not rigorously established.