arXiv cs.AI· Chao Xu, Maohua Li, Qirui Li, Yixuan Xu, Yanke Zhou, Yunhe Li, Cuifeng Shen, Hanlin Tang, Kan Liu, Tao Lan, Lin Qu, Shao-Qun Zhang·· 13 小时前AI 评分32
重新审视扩散 Transformer 中的跨层信息路由
Rethinking Cross-Layer Information Routing in Diffusion Transformers
AI 导读
研究对 Diffusion Transformers（DiTs）的跨层信息流做系统实证分析，指出传统残差相加存在前向幅值单调膨胀、反向梯度急剧衰减和块间冗余三个问题，并提出即插即用的残差替代方案 Diffusion-Adaptive Routing（DAR），对子层输出历史做可学习、随去噪时间步自适应的非增量聚合。
正文 · 原文
Authors:Chao Xu, Maohua Li, Qirui Li, Yixuan Xu, Yanke Zhou, Yunhe Li, Cuifeng Shen, Hanlin Tang, Kan Liu, Tao Lan, Lin Qu, Shao-Qun Zhang
Abstract:Diffusion Transformers (DiTs) have become a de facto backbone of modern visual generation, and nearly every major axis of their design -- tokenization, attention, conditioning, objectives, and latent autoencoders -- has been extensively revisited. The residual stream that governs how information accumulates across layers, however, has been directly inherited from the original Transformer. In this paper, we present a systematic empirical analysis of cross-layer information flow in DiTs, jointly along depth and denoising timestep, and identify three concrete symptoms of traditional residual addition, namely monotonic forward magnitude inflation, sharp backward gradient decay, and pronounced block-wise redundancy. Motivated by this diagnosis, we propose Diffusion-Adaptive Routing (DAR), a drop-in residual replacement that performs learnable, timestep-adaptive, and non-incremental aggregation over the history of sublayer outputs. Moreover, the proposed DAR is compatible with many modern Transformer enhancement methods, such as REPA. On ImageNet $256\times256$, DAR improves SiT-XL/2 by $2.11$ FID ($7.56$ vs. $9.67$) and matches the baseline's converged quality with $8.75\times$ fewer training iterations. Stacked on top of REPA, it yields a $2\times$ training acceleration in the early stage, suggesting cross-layer information routing as an underexplored design axis in diffusion modeling, one that operates orthogonally to existing representation-alignment objectives. Beyond pretraining, DAR can also be applied during the fine-tuning stage of large-scale T2I models and preserves high-frequency details during Distribution Matching Distillation. Code is available at this https URL.
| Comments: | NeurIPS 2026 Poster |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.20708 [cs.CV] |
| (or arXiv:2605.20708v4 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2605.20708 arXiv-issued DOI via DataCite |
Submission history
From: Maohua Li [view email]
[v1]
Wed, 20 May 2026 05:07:15 UTC (5,460 KB)
[v2]
Tue, 16 Jun 2026 17:19:08 UTC (5,460 KB)
[v3]
Wed, 30 Sep 2026 09:30:50 UTC (5,456 KB)
[v4]
Tue, 20 Oct 2026 08:27:54 UTC (5,448 KB)
来源：arXiv cs.AI · arxiv.org