arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2006.14548stat.MLcond-mat.dis-nncs.LGcs.NE
收藏

张量程序II:适用于任何体系结构的神经正切核

Tensor Programs II: Neural Tangent Kernel for Any Architecture

Microsoft Research AI(微软研究院AI)

Greg Yang

摘要:

我们证明了一个随机初始化的神经网络的 * 任何架构 * 有它的切线核(NTK)收敛到一个确定性的限制,网络宽度趋于无穷大。我们演示如何计算这个极限。在以前的文献中,对神经网络梯度的启发式研究通常假设前向传播中使用的每个权重矩阵与反向传播中使用的转置无关(Schoenholz et al. 2017)。这就是所谓的梯度独立假设(GIA)。我们确定了一个普遍满足的条件,我们称之为 * 简单GIA检查 *,这样基于GIA的NTK限值计算是正确的。相反,当简单GIA检查失败时,我们表明GIA可能会导致错误的答案。我们这里的材料以友好的方式呈现了Yang(2019 a)的NTK结果,并展示了用于理解广泛神经网络的 * 张量程序 * 技术。我们在https://github.com/thegregyang/NTK4A上提供了递归神经网络、Transformer和批量归一化的无限宽NTK的参考实现。

英文摘要:

We prove that a randomly initialized neural network of *any architecture* has its Tangent Kernel (NTK) converge to a deterministic limit, as the network widths tend to infinity. We demonstrate how to calculate this limit. In prior literature, the heuristic study of neural network gradients often assumes every weight matrix used in forward propagation is independent from its transpose used in backpropagation (Schoenholz et al. 2017). This is known as the *gradient independence assumption (GIA)*. We identify a commonly satisfied condition, which we call *Simple GIA Check*, such that the NTK limit calculation based on GIA is correct. Conversely, when Simple GIA Check fails, we show GIA can result in wrong answers. Our material here presents the NTK results of Yang (2019a) in a friendly manner and showcases the *tensor programs* technique for understanding wide neural networks. We provide reference implementations of infinite-width NTKs of recurrent neural network, transformer, and batch normalization at https://github.com/thegregyang/NTK4A.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑