Classifier accuracy versus input noise shows a sharp transition, and the transition width decreases as a power law (exponent about 0.27) with the number of model parameters for EfficientNet-type models.
Seidler \ and\ author M
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
physics.data-an 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Transition of AI Models in dependence of noise
Classifier accuracy versus input noise shows a sharp transition, and the transition width decreases as a power law (exponent about 0.27) with the number of model parameters for EfficientNet-type models.