Structured Initialization for Attention in Vision Transformers

arXiv:2404.01139v1 Announce Type: new
Abstract: The training of vision transformer (ViT) networks on small-scale datasets poses a significant challenge. By contrast, convolutional neural networks (CNNs) have an architectural inductive bias enabling them to perform well on such problems. In this paper, we argue that the architectural bias inherent to CNNs can be reinterpreted as an initialization bias within ViT. This insight is significant as it empowers ViTs to perform equally well on small-scale problems while maintaining their flexibility for large-scale …

abstract arxiv attention bias challenge cnns contrast convolutional neural networks cs.cv datasets enabling inductive networks neural networks paper scale small them training transformer transformers type vision vision transformers vit