arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1703.00560cs.LG
收藏

双层ReLU网络种群梯度的解析公式及其在收敛性和临界点分析中的应用

An Analytical Formula of Population Gradient for two-layered ReLU network and its Applications in Convergence and Critical Point Analysis

Facebook AI Research(Facebook人工智能研究院)

Yuandong Tian

摘要:

在本文中,我们探索了训练双层ReLU网络$g(\mathbf{x}; \mathbf{w}) = \sum_{j=1}^K σ(\mathbf{w}_j^T\mathbf{x})$的理论性质,其中中心$d$维球形高斯输入$\mathbf{x}$($σ$ =ReLU)。我们在$\mathbf{w}$上用梯度下降来训练我们的网络,以模仿具有相同架构和固定参数$\mathbf{w}^*$的教师网络的输出。我们表明,它的人口梯度有一个解析公式,导致有趣的理论分析的临界点和收敛行为。首先,我们证明了教师参数(“平面外”)所张成的超平面外的临界点不是孤立的,并形成流形,并描述了两种ReLU情况下的平面内无临界点区域。另一方面,如果权重被随机初始化,标准偏差上限由$O(ε/\sqrt{d})$限定,则至少以$(1-ε)/2$概率保证一个ReLU节点收敛到$\mathbf{w}^*$,这与经验实践一致。对于具有许多ReLU节点的网络,我们证明了权重初始化的无穷小扰动会导致收敛到$\mathbf{w}^*$(或其排列),这是物理学中称为自发破缺(SSB)的现象。我们假设ReLU激活没有独立性。仿真验证了我们的研究结果。

英文摘要:

In this paper, we explore theoretical properties of training a two-layered ReLU network $g(\mathbf{x}; \mathbf{w}) = \sum_{j=1}^K σ(\mathbf{w}_j^T\mathbf{x})$ with centered $d$-dimensional spherical Gaussian input $\mathbf{x}$ ($σ$=ReLU). We train our network with gradient descent on $\mathbf{w}$ to mimic the output of a teacher network with the same architecture and fixed parameters $\mathbf{w}^*$. We show that its population gradient has an analytical formula, leading to interesting theoretical analysis of critical points and convergence behaviors. First, we prove that critical points outside the hyperplane spanned by the teacher parameters ("out-of-plane") are not isolated and form manifolds, and characterize in-plane critical-point-free regions for two ReLU case. On the other hand, convergence to $\mathbf{w}^*$ for one ReLU node is guaranteed with at least $(1-ε)/2$ probability, if weights are initialized randomly with standard deviation upper-bounded by $O(ε/\sqrt{d})$, consistent with empirical practice. For network with many ReLU nodes, we prove that an infinitesimal perturbation of weight initialization results in convergence towards $\mathbf{w}^*$ (or its permutation), a phenomenon known as spontaneous symmetric-breaking (SSB) in physics. We assume no independence of ReLU activations. Simulation verifies our findings.

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。

↑