ANCRe：自适应神经连接重分配实现高效深度扩展
ANCRe: Adaptive Neural Connection Reassignment for Efficient Depth Scaling
AI 导读
arXiv 论文提出自适应神经连接重分配框架 ANCRe，从优化角度分析残差连接布局如何决定收敛行为，甚至导致收敛率的指数级差距。ANCRe 通过从数据中学习残差连接方式并自适应重分配，计算与内存开销不足 1%，在大语言模型预训练、扩散模型和深度 ResNets 上均实现更快收敛、更高性能与更好的深度利用效率。
正文 · 原文
Abstract:Scaling network depth has been a central driver behind the success of modern foundation models, yet recent investigations suggest that deep layers are often underutilized. This paper revisits the default mechanism for deepening neural networks, namely residual connections, from an optimization perspective. Rigorous analysis proves that the layout of residual connections can fundamentally shape convergence behavior, and even induces an exponential gap in convergence rates. Prompted by this insight, we introduce adaptive neural connection reassignment (ANCRe), a principled and lightweight framework that parameterizes and learns residual connectivities from the data. ANCRe adaptively reassigns residual connections with negligible computational and memory overhead ($<1\%$), while enabling more effective utilization of network depth. Extensive numerical tests across pre-training of large language models, diffusion models, and deep ResNets demonstrate consistently accelerated convergence, boosted performance, and enhanced depth efficiency over conventional residual connections.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.09009 [cs.LG] |
| (or arXiv:2602.09009v5 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2602.09009 arXiv-issued DOI via DataCite |
|
| Journal reference: | Advances in Neural Information Processing Systems, 2026 |
Submission history
From: Yilang Zhang [view email]
[v1]
Mon, 9 Feb 2026 18:54:18 UTC (5,679 KB)
[v2]
Thu, 1 Oct 2026 21:55:31 UTC (5,686 KB)
[v3]
Tue, 6 Oct 2026 19:52:15 UTC (5,686 KB)
[v4]
Tue, 13 Oct 2026 21:55:31 UTC (5,689 KB)
[v5]
Wed, 11 Nov 2026 22:03:55 UTC (5,689 KB)
来源：arXiv cs.AI · arxiv.org