DSBP claims faster, flatter training by projecting gradients onto principal activation eigenvectors, with reported gains over SAM, LoRA, and MAML, but the math is internally inconsistent and the results are not reproducible from the paper.
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
-
Dynamic Spectral Backpropagation for Efficient Neural Network Training
DSBP claims faster, flatter training by projecting gradients onto principal activation eigenvectors, with reported gains over SAM, LoRA, and MAML, but the math is internally inconsistent and the results are not reproducible from the paper.