Diffusion models suffer representation degradation at high noise due to recoverability mismatch; ERD mitigates this by dynamic optimization reallocation, accelerating convergence across backbones.
Wide neural networks of any depth evolve as linear models under gradient descent
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Elucidating Representation Degradation Problem in Diffusion Model Training
Diffusion models suffer representation degradation at high noise due to recoverability mismatch; ERD mitigates this by dynamic optimization reallocation, accelerating convergence across backbones.