We consider the task of estimating a low-rank matrix from non-linear and noisy observations. We prove a strong universality result showing that Bayes-optimal performances are characterized by an equivalent Gaussian model with an effective prior, whose parameters are entirely determined by an expansion of the non-linear function. In particular, we show that to reconstruct the signal accurately, one requires a signal-to-noise ratio growing as $N^{\frac 12 (1-1/k_F)}$, where $k_F$ is the first non-zero Fisher information coefficient of the function. We provide asymptotic characterization for the minimal achievable mean squared error (MMSE) and an approximate message-passing algorithm that reaches the MMSE under conditions analogous to the linear version of the problem. We also provide asymptotic errors achieved by methods such as principal component analysis combined with Bayesian denoising, and compare them with Bayes-optimal MMSE.

从非线性和含噪声观测中估计一个低秩矩阵的任务中，我们证明了一个强普适性结果，表明贝叶斯最优性能可由一个等效的高斯模型表示，其先验参数完全由非线性函数的展开所确定。特别地，我们展示了为了准确重建信号，需要一个随着$ N^{rac 12(1-1/k_F)}$增长的信噪比，其中$k_F$是函数的第一个非零Fisher信息系数。我们提供了最小可实现均方误差（MMSE）的渐近特征及一个类似于问题的线性版本的条件下能达到MMSE的近似传递算法。我们还提供了方法如主成分分析与贝叶斯去噪等的渐近误差，并将其与贝叶斯最优MMSE进行了比较。

非线性低秩矩阵估计的基本限制