arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2405.16630stat.MLcs.AIcs.LGmath.PRphysics.data-an
收藏

深度弱非线性网络的贝叶斯推断

Bayesian Inference with Deep Weakly Nonlinear Networks

Princeton ORFE(普林斯顿大学运筹学与金融工程系);MIT Center for Theoretical Physics(麻省理工学院理论物理中心);Google Quantum AI(谷歌量子人工智能)

Boris Hanin, Alexander Zlokapa

AI总结:

本文证明深度弱非线性网络的贝叶斯推断在特定大尺度条件下可解,并揭示其与核方法的联系及深度对特征学习和泛化的益处。

摘要:

我们在物理学严谨性层面上证明,对于具有全连接神经网络和形如 $\phi(t) = t + \psi t^3/L$ 的整形非线性的贝叶斯推断,在训练数据点数量 $P$、输入维度 $N_0$、网络层宽度 $N$ 和网络深度 $L$ 同时很大的情况下,是(微扰意义上)可解的。我们的结果在数据的弱假设下成立;主要约束是 $P < N_0$。我们提供了计算模型证据和后验的技术,可达到 $1/N$ 的任意阶以及任意温度。我们报告了以下一阶计算的结果:\n1. 当宽度 $N$ 远大于深度 $L$ 和训练集大小 $P$ 时,神经网络贝叶斯推断与使用核的贝叶斯推断一致。$\psi$ 的值决定了训练数据在特征映射下被隐式嵌入的球面、双曲面或平面的曲率。\n2. 当 $LP/N$ 是一个小常数时,神经网络贝叶斯推断偏离核机制。在零温度下,神经网络贝叶斯推断等价于使用数据相关核的贝叶斯推断,而 $LP/N$ 作为一个有效深度,控制着特征学习的程度。\n3. 在深度线性网络($\psi=0$)和含噪数据的受限情况下,我们展示了一个简单数据模型,其证据和泛化误差在零温度下达到最优。随着 $LP/N$ 的增加,证据和泛化性能都进一步改善,展示了深度在良性过拟合中的益处。

英文摘要:

We show at a physics level of rigor that Bayesian inference with a fully connected neural network and a shaped nonlinearity of the form $ϕ(t) = t + ψt^3/L$ is (perturbatively) solvable in the regime where the number of training datapoints $P$ , the input dimension $N_0$, the network layer widths $N$, and the network depth $L$ are simultaneously large. Our results hold with weak assumptions on the data; the main constraint is that $P < N_0$. We provide techniques to compute the model evidence and posterior to arbitrary order in $1/N$ and at arbitrary temperature. We report the following results from the first-order computation: 1. When the width $N$ is much larger than the depth $L$ and training set size $P$, neural network Bayesian inference coincides with Bayesian inference using a kernel. The value of $ψ$ determines the curvature of a sphere, hyperbola, or plane into which the training data is implicitly embedded under the feature map. 2. When $LP/N$ is a small constant, neural network Bayesian inference departs from the kernel regime. At zero temperature, neural network Bayesian inference is equivalent to Bayesian inference using a data-dependent kernel, and $LP/N$ serves as an effective depth that controls the extent of feature learning. 3. In the restricted case of deep linear networks ($ψ=0$) and noisy data, we show a simple data model for which evidence and generalization error are optimal at zero temperature. As $LP/N$ increases, both evidence and generalization further improve, demonstrating the benefit of depth in benign overfitting.

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。

↑