ONG prunes neural network weights at initialization using NMF reconstruction error as an importance score, then strictly preserves the mask throughout training.
On the lattices of exact and weakly exact structures
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We initiate in this article the study of weakly exact structures, a generalization of Quillen exact structures. We introduce weak counterparts of one-sided exact structures and show that a left and a right weakly exact structure generate a weakly exact structure. We further define weakly extriangulated structures on an additive category $\mathcal{A}$ and characterize weakly exact structures among them. We investigate when these structures on $\mathcal{A}$ form lattices. We prove that the lattice of substructures of a weakly extriangulated structure is isomorphic to the lattice of topologizing subcategories of a certain functor category. In the idempotent complete case, we characterize the lattice of all weakly exact structures and we prove the existence of a unique maximal weakly exact structure. We study in detail the situation when $\mathcal{A}$ is additively finite, giving a module-theoretic characterization of closed sub-bifunctors of $\mbox{Ext}^1$ among all additive sub-bifunctors.
citation-role summary
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
REJECT 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
ONG: One-Shot NMF-based Gradient Masking for Efficient Model Sparsification
ONG prunes neural network weights at initialization using NMF reconstruction error as an importance score, then strictly preserves the mask throughout training.