Extreme neural network sparsification leads to catastrophic local interpretability collapse despite stable global metrics, with dead neuron rates up to 90% on complex datasets.
Title resolution pending
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
-
Fundamental Limits of Neural Network Sparsification: Evidence from Catastrophic Interpretability Collapse
Extreme neural network sparsification leads to catastrophic local interpretability collapse despite stable global metrics, with dead neuron rates up to 90% on complex datasets.