← 研踪 PaperTrailarXiv:2611.25624 · v1
Deviations from the NTK Regime in CNNs under Joint Depth-Width Scaling
首次提交 2026-11-19 15:09
版本更新 2026-11-19 15:09
雷达观测 2026-11-20 19:58
摘要
The concentration and training-time constancy of the neural tangent kernel (NTK) depend on the interplay between network width, depth, and initialization. While joint depth-width scaling has been studied in fully-connected, residual, and recurrent networks, its effects in convolutional neural networks (CNNs) remain less understood. We investigate these effects numerically in CNNs, examining NTK dispersion at initialization and kernel rotation during training. Let depth be the number of convolutional layers, and width the number of channels per layer. We find that NTK dispersion increases with in the critical and chaotic initialization regimes, but remains small in the ordered regime. At critical initialization, dispersion curves approximately collapse as a function of , suggesting a nontrivial joint scaling regime. During training, kernel rotation generally increases with , even in the ordered regime where initial dispersion is small. Together, these results suggest that may govern NTK deviations from the infinite-width limit in certain regimes, and that NTK concentration at initialization and stability during training exhibit different scaling behaviors.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 15:09 · 最新观测版本
Deviations from the NTK Regime in CNNs under Joint Depth-Width Scaling
The concentration and training-time constancy of the neural tangent kernel (NTK) depend on the interplay between network width, depth, and initialization. While joint depth-width scaling has been studied in fully-connected, residual, and recurrent networks, its effects in convolutional neural networks (CNNs) remain less understood. We investigate these effects numerically in CNNs, examining NTK dispersion at initialization and kernel rotation during training. Let depth be the number of convolutional layers, and width the number of channels per layer. We find that NTK dispersion increases with in the critical and chaotic initialization regimes, but remains small in the ordered regime. At critical initialization, dispersion curves approximately collapse as a function of , suggesting a nontrivial joint scaling regime. During training, kernel rotation generally increases with , even in the ordered regime where initial dispersion is small. Together, these results suggest that may govern NTK deviations from the infinite-width limit in certain regimes, and that NTK concentration at initialization and stability during training exhibit different scaling behaviors.
查看此版本解读 →查看此版本原文 ↗