Salvato in:
Dettagli Bibliografici
Autori principali: Ji, Yiping, Martens, James, Zheng, Jianqiao, Zhou, Ziqin, Moghadam, Peyman, Zhang, Xinyu, Saratchandran, Hemanth, Lucey, Simon
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2510.00345
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916982069133312
author Ji, Yiping
Martens, James
Zheng, Jianqiao
Zhou, Ziqin
Moghadam, Peyman
Zhang, Xinyu
Saratchandran, Hemanth
Lucey, Simon
author_facet Ji, Yiping
Martens, James
Zheng, Jianqiao
Zhou, Ziqin
Moghadam, Peyman
Zhang, Xinyu
Saratchandran, Hemanth
Lucey, Simon
contents Transformers have achieved remarkable success across a wide range of applications, a feat often attributed to their scalability. Yet training them without skip (residual) connections remains notoriously difficult. While skips stabilize optimization, they also disrupt the hierarchical structure of representations, raising the long-standing question of whether transformers can be trained efficiently without them. In this work, we address this problem by analyzing the Jacobian of a skipless transformer block, showing why skips improve conditioning and revealing that their stabilization benefits can be recovered through a principled initialization strategy. Building on this insight, we introduce the first method that enables stable and efficient training of skipless transformers without altering the standard architecture. We validate our approach on Vision Transformers (ViTs) in both supervised and self-supervised settings, demonstrating that skipless ViTs trained with our initialization overcome the usual optimization barriers, learn richer hierarchical representations, and outperform strong baselines, that incorporate skip connections, on dense prediction benchmarks. These results show that skip connections are not a fundamental requirement for training ViTs and open new avenues for hierarchical representation learning in vision models.
format Preprint
id arxiv_https___arxiv_org_abs_2510_00345
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cutting the Skip: Training Residual-Free Transformers
Ji, Yiping
Martens, James
Zheng, Jianqiao
Zhou, Ziqin
Moghadam, Peyman
Zhang, Xinyu
Saratchandran, Hemanth
Lucey, Simon
Machine Learning
Transformers have achieved remarkable success across a wide range of applications, a feat often attributed to their scalability. Yet training them without skip (residual) connections remains notoriously difficult. While skips stabilize optimization, they also disrupt the hierarchical structure of representations, raising the long-standing question of whether transformers can be trained efficiently without them. In this work, we address this problem by analyzing the Jacobian of a skipless transformer block, showing why skips improve conditioning and revealing that their stabilization benefits can be recovered through a principled initialization strategy. Building on this insight, we introduce the first method that enables stable and efficient training of skipless transformers without altering the standard architecture. We validate our approach on Vision Transformers (ViTs) in both supervised and self-supervised settings, demonstrating that skipless ViTs trained with our initialization overcome the usual optimization barriers, learn richer hierarchical representations, and outperform strong baselines, that incorporate skip connections, on dense prediction benchmarks. These results show that skip connections are not a fundamental requirement for training ViTs and open new avenues for hierarchical representation learning in vision models.
title Cutting the Skip: Training Residual-Free Transformers
topic Machine Learning
url https://arxiv.org/abs/2510.00345