arxiv:2409.19606

Hyper-Connections

Published on Sep 29

· Submitted by

Taoer on Oct 1

#3 Paper of the day

Upvote

Authors:

Defa Zhu ,

Hongzhi Huang ,

Zihao Huang ,

Yutao Zeng ,

Yunyao Mao ,

Banggu Wu ,

Qiyang Min ,

Xun Zhou

Abstract

We present hyper-connections, a simple yet effective method that can serve as an alternative to residual connections. This approach specifically addresses common drawbacks observed in residual connection variants, such as the seesaw effect between gradient vanishing and representation collapse. Theoretically, hyper-connections allow the network to adjust the strength of connections between features at different depths and dynamically rearrange layers. We conduct experiments focusing on the pre-training of large language models, including dense and sparse models, where hyper-connections show significant performance improvements over residual connections. Additional experiments conducted on vision tasks also demonstrate similar improvements. We anticipate that this method will be broadly applicable and beneficial across a wide range of AI problems.

View arXiv page View PDF Add to collection

Community

Taoer

Paper author Paper submitter Oct 1

•

edited Oct 1

TL;DR: We present hyper-connections, a simple yet effective method that can serve as an alternative to residual connections. It addresses issues like gradient vanishing and representation collapse by allowing networks to adjust connection strengths and rearrange layers dynamically. Experiments with LLMs Pre-training and vision tasks show significant performance improvements, suggesting broad applicability in AI.