NTK regression requires exponentially more samples than the minimax rate of depth-L ReLU networks whenever a target's Fourier complexity decouples from its architectural complexity, as on the depth-L sawtooth.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
A Function-Space Dichotomy for Compositional Learning: Exponential Sub-Optimality of the Neural Tangent Kernel
NTK regression requires exponentially more samples than the minimax rate of depth-L ReLU networks whenever a target's Fourier complexity decouples from its architectural complexity, as on the depth-L sawtooth.