Patch-level Routing in Mixture-of-Experts is Provably Sample-efficient for Convolutional Neural Networks
Patch-level Routing in Mixture-of-Experts is Provably Sample-efficient for Convolutional Neural Networks
复制标题
DOI：
10.48550/arxiv.2306.04073
复制
发表时间：
2023-06
期刊：
影响因子：
--
通讯作者：
Mohammed Nowaz Rabbani Chowdhury;Shuai Zhang;M. Wang;Sijia Liu;Pin-Yu Chen
中科院分区：
文献类型：
--
作者：
Mohammed Nowaz Rabbani Chowdhury;Shuai Zhang;M. Wang;Sijia Liu;Pin-Yu Chen
在深度学习中，专家（MOE）的混合物以每样本或按the的基础激活一个或几个专家（子网络），从而大大降低计算。最近提议的\下划线{p} ATCH级路由\下划线{Moe}（PMOE）将每个输入分为$ n $ patches（或代币），并通过优先的路由将$ l $ patches（$ l \ ll n $）发送给每个专家。 PMOE在降低培训和推理成本方面取得了巨大的经验成功，同时保持了测试准确性。但是，PMOE和将军的理论解释仍然难以捉摸。通过两层卷积神经网络（CNN）的混合，专注于监督分类任务，我们首次表明，PMOE可证明PMOE可以通过$ N/L $ and Perfors的多项序列中的一个因素来减少所需的培训样本数量，以实现理想的概括（称为样品复杂性）。优势是由歧视性路由属性引起的，这在理论和实践中都是合理的，即PMOE路由器可以过滤标签 - iRretrevant贴片并将相似的类别歧视贴剂路由到同一专家。我们对MNIST，CIFAR-10和CELEBA的实验结果支持我们对PMOE概括的理论发现，并表明PMOE可以避免学习虚假的相关性。
In deep learning, mixture-of-experts (MoE) activates one or few experts (sub-networks) on a per-sample or per-token basis, resulting in significant computation reduction. The recently proposed \underline{p}atch-level routing in \underline{MoE} (pMoE) divides each input into $n$ patches (or tokens) and sends $l$ patches ($l\ll n$) to each expert through prioritized routing. pMoE has demonstrated great empirical success in reducing training and inference costs while maintaining test accuracy. However, the theoretical explanation of pMoE and the general MoE remains elusive. Focusing on a supervised classification task using a mixture of two-layer convolutional neural networks (CNNs), we show for the first time that pMoE provably reduces the required number of training samples to achieve desirable generalization (referred to as the sample complexity) by a factor in the polynomial order of $n/l$, and outperforms its single-expert counterpart of the same or even larger capacity. The advantage results from the discriminative routing property, which is justified in both theory and practice that pMoE routers can filter label-irrelevant patches and route similar class-discriminative patches to the same expert. Our experimental results on MNIST, CIFAR-10, and CelebA support our theoretical findings on pMoE's generalization and show that pMoE can avoid learning spurious correlations.