Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userSpectral norm of random tensors
103
0
0.0
(
0
)
Added by
Ryota Tomioka
Publication date
2014
fields
Mathematical Statistics
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
We show that the spectral norm of a random $n_1times n_2times cdots times n_K$ tensor (or higher-order array) scales as $Oleft(sqrt{(sum_{k=1}^{K}n_k)log(K)}right)$ under some sub-Gaussian assumption on the entries. The proof is based on a covering number argument. Since the spectral norm is dual to the tensor nuclear norm (the tightest convex relaxation of the set of rank one tensors), the bound implies that the convex relaxation yields sample complexity that is linear in (the sum of) the number of dimensions, which is much smaller than other recently proposed convex relaxations of tensor rank that use unfolding.
rate research
Read More
Consider the classical supervised learning problem: we are given data $(y_i,{boldsymbol x}_i)$, $ile n$, with $y_i$ a response and ${boldsymbol x}_iin {mathcal X}$ a covariates vector, and try to learn a model $f:{mathcal X}to{mathbb R}$ to predict future responses. Random features methods map the covariates vector ${boldsymbol x}_i$ to a point ${boldsymbol phi}({boldsymbol x}_i)$ in a higher dimensional space ${mathbb R}^N$, via a random featurization map ${boldsymbol phi}$. We study the use of random features methods in conjunction with ridge regression in the feature space ${mathbb R}^N$. This can be viewed as a finite-dimensional approximation of kernel ridge regression (KRR), or as a stylized model for neural networks in the so called lazy training regime. We define a class of problems satisfying certain spectral conditions on the underlying kernels, and a hypercontractivity assumption on the associated eigenfunctions. These conditions are verified by classical high-dimensional examples. Under these conditions, we prove a sharp characterization of the error of random features ridge regression. In particular, we address two fundamental questions: $(1)$~What is the generalization error of KRR? $(2)$~How big $N$ should be for the random features approximation to achieve the same error as KRR? In this setting, we prove that KRR is well approximated by a projection onto the top $ell$ eigenfunctions of the kernel, where $ell$ depends on the sample size $n$. We show that the test error of random features ridge regression is dominated by its approximation error and is larger than the error of KRR as long as $Nle n^{1-delta}$ for some $delta>0$. We characterize this gap. For $Nge n^{1+delta}$, random features achieve the same error as the corresponding KRR, and further increasing $N$ does not lead to a significant change in test error.
We investigate the construction of early stopping rules in the nonparametric regression problem where iterative learning algorithms are used and the optimal iteration number is unknown. More precisely, we study the discrepancy principle, as well as modifications based on smoothed residuals, for kernelized spectral filter learning algorithms including gradient descent. Our main theoretical bounds are oracle inequalities established for the empirical estimation error (fixed design), and for the prediction error (random design). From these finite-sample bounds it follows that the classical discrepancy principle is statistically adaptive for slow rates occurring in the hard learning scenario, while the smoothed discrepancy principles are adaptive over ranges of faster rates (resp. higher smoothness parameters). Our approach relies on deviation inequalities for the stopping rules in the fixed design setting, combined with change-of-norm arguments to deal with the random design setting.
In data science, it is often required to estimate dependencies between different data sources. These dependencies are typically calculated using Pearsons correlation, distance correlation, and/or mutual information. However, none of these measures satisfy all the Grangers axioms for an ideal measure. One such ideal measure, proposed by Granger himself, calculates the Bhattacharyya distance between the joint probability density function (pdf) and the product of marginal pdfs. We call this measure the mutual dependence. However, to date this measure has not been directly computable from data. In this paper, we use our recently introduced maximum likelihood non-parametric estimator for band-limited pdfs, to compute the mutual dependence directly from the data. We construct the estimator of mutual dependence and compare its performance to standard measures (Pearsons and distance correlation) for different known pdfs by computing convergence rates, computational complexity, and the ability to capture nonlinear dependencies. Our mutual dependence estimator requires fewer samples to converge to theoretical values, is faster to compute, and captures more complex dependencies than standard measures.
The classical setting of community detection consists of networks exhibiting a clustered structure. To more accurately model real systems we consider a class of networks (i) whose edges may carry labels and (ii) which may lack a clustered structure. Specifically we assume that nodes possess latent attributes drawn from a general compact space and edges between two nodes are randomly generated and labeled according to some unknown distribution as a function of their latent attributes. Our goal is then to infer the edge label distributions from a partially observed network. We propose a computationally efficient spectral algorithm and show it allows for asymptotically correct inference when the average node degree could be as low as logarithmic in the total number of nodes. Conversely, if the average node degree is below a specific constant threshold, we show that no algorithm can achieve better inference than guessing without using the observations. As a byproduct of our analysis, we show that our model provides a general procedure to construct random graph models with a spectrum asymptotic to a pre-specified eigenvalue distribution such as a power-law distribution.
Consider a random vector $mathbf{y}=mathbf{Sigma}^{1/2}mathbf{x}$, where the $p$ elements of the vector $mathbf{x}$ are i.i.d. real-valued random variables with zero mean and finite fourth moment, and $mathbf{Sigma}^{1/2}$ is a deterministic $ptimes p$ matrix such that the spectral norm of the population correlation matrix $mathbf{R}$ of $mathbf{y}$ is uniformly bounded. In this paper, we find that the log determinant of the sample correlation matrix $hat{mathbf{R}}$ based on a sample of size $n$ from the distribution of $mathbf{y}$ satisfies a CLT (central limit theorem) for $p/nto gammain (0, 1]$ and $pleq n$. Explicit formulas for the asymptotic mean and variance are provided. In case the mean of $mathbf{y}$ is unknown, we show that after recentering by the empirical mean the obtained CLT holds with a shift in the asymptotic mean. This result is of independent interest in both large dimensional random matrix theory and high-dimensional statistical literature of large sample correlation matrices for non-normal data. At last, the obtained findings are applied for testing of uncorrelatedness of $p$ random variables. Surprisingly, in the null case $mathbf{R}=mathbf{I}$, the test statistic becomes completely pivotal and the extensive simulations show that the obtained CLT also holds if the moments of order four do not exist at all, which conjectures a promising and robust test statistic for heavy-tailed high-dimensional data.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
