When Training-Free NAS Meets Vision Transformer: A Neural Tangent Kernel Perspective

TL;DR

This paper explores the limitations of Neural Tangent Kernel-based metrics for vision transformer NAS and proposes ViNTK, a Fourier feature-enhanced NTK, to improve search efficiency and performance.

Contribution

It introduces ViNTK, a novel NTK extension that captures high-frequency features, addressing previous limitations in ViT neural architecture search.

Findings

ViNTK significantly reduces NAS search costs.

ViNTK maintains competitive accuracy on classification and segmentation.

NTK's low-frequency bias limits its effectiveness for ViT NAS.

Abstract

This paper investigates the Neural Tangent Kernel (NTK) to search vision transformers without training. In contrast with the previous observation that NTK-based metrics can effectively predict CNNs performance at initialization, we empirically show their inefficacy in the ViT search space. We hypothesize that the fundamental feature learning preference within ViT contributes to the ineffectiveness of applying NTK to NAS for ViT. We both theoretically and empirically validate that NTK essentially estimates the ability of neural networks that learn low-frequency signals, completely ignoring the impact of high-frequency signals in feature learning. To address this limitation, we propose a new method called ViNTK that generalizes the standard NTK to the high-frequency domain by integrating the Fourier features from inputs. Experiments with multiple ViT search spaces on image classification and semantic segmentation tasks show that our method can significantly speed up search costs over prior state-of-the-art NAS for ViT while maintaining similar performance on searched architectures.