MLPs at the EOC: Concentration of the NTK

TL;DR

This paper analyzes the concentration of the Neural Tangent Kernel (NTK) in multilayer perceptrons initialized at the Edge Of Chaos, showing finite-width conditions for the NTK to approximate its infinite-width limit without relying on gradient independence.

Contribution

It proves that the NTK concentrates around its limit at finite width for MLPs with specific activation functions, without assuming linear overparameterization, and identifies quadratic hidden layer width growth as sufficient.

Findings

NTK concentrates around its infinite-width limit at finite width.

Activation functions with certain parameters improve NTK concentration.

Quadratic growth in hidden layer widths ensures accurate NTK approximation.

Abstract

We study the concentration of the Neural Tangent Kernel (NTK) $K_\theta : \mathbb{R}^{m_0} \times \mathbb{R}^{m_0} \to \mathbb{R}^{m_l \times m_l}$ of $l$-layer Multilayer Perceptrons (MLPs) $N : \mathbb{R}^{m_0} \times \Theta \to \mathbb{R}^{m_l}$ equipped with activation functions $\phi(s) = a s + b \vert s \vert$ for some $a,b \in \mathbb{R}$ with the parameter $\theta \in \Theta$ being initialized at the Edge Of Chaos (EOC). Without relying on the gradient independence assumption that has only been shown to hold asymptotically in the infinitely wide limit, we prove that an approximate version of gradient independence holds at finite width. Showing that the NTK entries $K_\theta(x_{i_1},x_{i_2})$ for $i_1,i_2 \in [1:n]$ over a dataset $\{x_1,\cdots,x_n\} \subset \mathbb{R}^{m_0}$ concentrate simultaneously via maximal inequalities, we prove that the NTK matrix $K(\theta) = [\frac{1}{n} K_\theta(x_{i_1},x_{i_2}) : i_1,i_2 \in [1:n]] \in \mathbb{R}^{nm_l \times nm_l}$ concentrates around its infinitely wide limit $\overset{\scriptscriptstyle\infty}{K} \in \mathbb{R}^{nm_l \times nm_l}$ without the need for linear overparameterization. Our results imply that in order to accurately approximate the limit, hidden layer widths have to grow quadratically as $m_k = k^2 m$ for some $m \in \mathbb{N}+1$ for sufficient concentration. For such MLPs, we obtain the concentration bound $\mathbb{P}( \Vert K(\theta) - \overset{\scriptscriptstyle\infty}{K} \Vert \leq O((\Delta_\phi^{-2} + m_l^{\frac{1}{2}} l) \kappa_\phi^2 m^{-\frac{1}{2}})) \geq 1-O(m^{-1})$ modulo logarithmic terms, where we denoted $\Delta_\phi = \frac{b^2}{a^2+b^2}$ and $\kappa_\phi = \frac{\vert a \vert + \vert b \vert}{\sqrt{a^2 + b^2}}$. This reveals in particular that the absolute value ($\Delta_\phi=1$, $\kappa_\phi=1$) beats the ReLU ($\Delta_\phi=\frac{1}{2}$, $\kappa_\phi=\sqrt{2}$) in terms of the concentration of the NTK.