When dealing with electro or magnetoencephalography records, many supervised prediction tasks are solved by working with covariance matrices to summarize the signals. Learning with these matrices requires using Riemanian geometry to account for their structure. In this paper, we propose a new method to deal with distributions of covariance matrices and demonstrate its computational efficiency on M/EEG multivariate time series. More specifically, we define a Sliced-Wasserstein distance between measures of symmetric positive definite matrices that comes with strong theoretical guarantees. Then, we take advantage of its properties and kernel methods to apply this distance to brain-age prediction from MEG data and compare it to state-of-the-art algorithms based on Riemannian geometry. Finally, we show that it is an efficient surrogate to the Wasserstein distance in domain adaptation for Brain Computer Interface applications.
相關內容
The automatic classification of 3D medical data is memory-intensive. Also, variations in the number of slices between samples is common. Naive solutions such as subsampling can solve these problems, but at the cost of potentially eliminating relevant diagnosis information. Transformers have shown promising performance for sequential data analysis. However, their application for long-sequences is data, computationally, and memory demanding. In this paper, we propose an end-to-end Transformer-based framework that allows to classify volumetric data of variable length in an efficient fashion. Particularly, by randomizing the input slice-wise resolution during training, we enhance the capacity of the learnable positional embedding assigned to each volume slice. Consequently, the accumulated positional information in each positional embedding can be generalized to the neighbouring slices, even for high resolution volumes at the test time. By doing so, the model will be more robust to variable volume length and amenable to different computational budgets. We evaluated the proposed approach in retinal OCT volume classification and achieved 21.96% average improvement in balanced accuracy on a 9-class diagnostic task, compared to state-of-the-art video transformers. Our findings show that varying the slice-wise resolution of the input during training results in more informative volume representation as compared to training with fixed number of slices per volume. Our code is available at: //github.com/marziehoghbaie/VLFAT.
Consider an empirical measure $\mathbb{P}_n$ induced by $n$ iid samples from a $d$-dimensional $K$-subgaussian distribution $\mathbb{P}$ and let $\gamma = \mathcal{N}(0,\sigma^2 I_d)$ be the isotropic Gaussian measure. We study the speed of convergence of the smoothed Wasserstein distance $W_2(\mathbb{P}_n * \gamma, \mathbb{P}*\gamma) = n^{-\alpha + o(1)}$ with $*$ being the convolution of measures. For $K<\sigma$ and in any dimension $d\ge 1$ we show that $\alpha = {1\over2}$. For $K>\sigma$ in dimension $d=1$ we show that the rate is slower and is given by $\alpha = {(\sigma^2 + K^2)^2\over 4 (\sigma^4 + K^4)} < 1/2$. This resolves several open problems in \cite{goldfeld2020convergence}, and in particular precisely identifies the amount of smoothing $\sigma$ needed to obtain a parametric rate. In addition, we also establish that $D_{KL}(\mathbb{P}_n * \gamma \|\mathbb{P}*\gamma)$ has rate $O(1/n)$ for $K<\sigma$ but only slows down to $O({(\log n)^{d+1}\over n})$ for $K>\sigma$. The surprising difference of the behavior of $W_2^2$ and KL implies the failure of $T_{2}$-transportation inequality when $\sigma < K$. Consequently, the requirement $K<\sigma$ is necessary for validity of the log-Sobolev inequality (LSI) for the Gaussian mixture $\mathbb{P} * \mathcal{N}(0, \sigma^{2})$, closing an open problem in \cite{wang2016functional}, who established the LSI under precisely this condition.
Distribution data refers to a data set where each sample is represented as a probability distribution, a subject area receiving burgeoning interest in the field of statistics. Although several studies have developed distribution-to-distribution regression models for univariate variables, the multivariate scenario remains under-explored due to technical complexities. In this study, we introduce models for regression from one Gaussian distribution to another, utilizing the Wasserstein metric. These models are constructed using the geometry of the Wasserstein space, which enables the transformation of Gaussian distributions into components of a linear matrix space. Owing to their linear regression frameworks, our models are intuitively understandable, and their implementation is simplified because of the optimal transport problem's analytical solution between Gaussian distributions. We also explore a generalization of our models to encompass non-Gaussian scenarios. We establish the convergence rates of in-sample prediction errors for the empirical risk minimizations in our models. In comparative simulation experiments, our models demonstrate superior performance over a simpler alternative method that transforms Gaussian distributions into matrices. We present an application of our methodology using weather data for illustration purposes.
Efficiently pricing multi-asset options is a challenging problem in quantitative finance. When the characteristic function is available, Fourier-based methods are competitive compared to alternative techniques because the integrand in the frequency space often has a higher regularity than that in the physical space. However, when designing a numerical quadrature method for most Fourier pricing approaches, two key aspects affecting the numerical complexity should be carefully considered: (i) the choice of damping parameters that ensure integrability and control the regularity class of the integrand and (ii) the effective treatment of high dimensionality. We propose an efficient numerical method for pricing European multi-asset options based on two complementary ideas to address these challenges. First, we smooth the Fourier integrand via optimized choice of damping parameters based on a proposed optimization rule. Second, we employ sparsification and dimension-adaptivity techniques to accelerate the convergence of the quadrature in high dimensions. The extensive numerical study on basket and rainbow options under the multivariate geometric Brownian motion and some L\'evy models demonstrates the advantages of adaptivity and the damping rule on the numerical complexity of quadrature methods. Moreover, the approach achieves substantial computational gains compared to the Monte Carlo method.
In [Sau11,SPW13], Saunderson, Parrilo and Willsky asked the following elegant geometric question: what is the largest $m= m(d)$ such that there is an ellipsoid in $\mathbb{R}^d$ that passes through $v_1, v_2, \ldots, v_m$ with high probability when the $v_i$s are chosen independently from the standard Gaussian distribution $N(0,I_{d})$. The existence of such an ellipsoid is equivalent to the existence of a positive semidefinite matrix $X$ such that $v_i^{\top}X v_i =1$ for every $1 \leq i \leq m$ - a natural example of a random semidefinite program. SPW conjectured that $m= (1-o(1)) d^2/4$ with high probability. Very recently, Potechin, Turner, Venkat and Wein and Kane and Diakonikolas proved that $m \geq d^2/\log^{O(1)}(d)$ via certain explicit constructions. In this work, we give a substantially tighter analysis of their construction to prove that $m \geq d^2/C$ for an absolute constant $C>0$. This resolves one direction of the SPW conjecture up to a constant. Our analysis proceeds via the method of Graphical Matrix Decomposition that has recently been used to analyze correlated random matrices arising in various areas [BHK+19]. Our key new technical tool is a refined method to prove singular value upper bounds on certain correlated random matrices that are tight up to absolute dimension-independent constants. In contrast, all previous methods that analyze such matrices lose logarithmic factors in the dimension.
Given a family of nearly commuting symmetric matrices, we consider the task of computing an orthogonal matrix that nearly diagonalizes every matrix in the family. In this paper, we propose and analyze randomized joint diagonalization (RJD) for performing this task. RJD applies a standard eigenvalue solver to random linear combinations of the matrices. Unlike existing optimization-based methods, RJD is simple to implement and leverages existing high-quality linear algebra software packages. Our main novel contribution is to prove robust recovery: Given a family that is $\epsilon$-near to a commuting family, RJD jointly diagonalizes this family, with high probability, up to an error of norm O($\epsilon$). No other existing method is known to enjoy such a universal robust recovery guarantee. We also discuss how the algorithm can be further improved by deflation techniques and demonstrate its state-of-the-art performance by numerical experiments with synthetic and real-world data.
Applications of the ensemble Kalman filter to high-dimensional problems are feasible only with small ensembles. This necessitates a kind of regularization of the analysis (observation update) problem. We propose a regularization technique based on a new non-stationary, non-parametric spatial model on the sphere. The model termed the Locally Stationary Convolution Model is a constrained version of the general Gaussian process convolution model. The constraints on the location-dependent convolution kernel include local isotropy, positive definiteness as a function of distance, and smoothness as a function of location. The model allows for a rigorous definition of the local spectrum, which is required to be a smooth function of spatial wavenumber. We propose and test an ensemble filter in which prior covariances are postulated to obey the Locally Stationary Convolution Model. The model is estimated online in a two-stage procedure. First, ensemble perturbations are bandpass filtered in several wavenumber bands to extract aggregated local spatial spectra. Second, a neural network recovers the local spectra from sample variances of the filtered fields. In simulation experiments, the new filter was capable of outperforming several existing techniques. With small to moderate ensemble sizes, the improvement was substantial.
Correlation matrices are an essential tool for investigating the dependency structures of random vectors or comparing them. We introduce an approach for testing a variety of null hypotheses that can be formulated based upon the correlation matrix. Examples cover MANOVA-type hypothesis of equal correlation matrices as well as testing for special correlation structures such as, e.g., sphericity. Apart from existing fourth moments, our approach requires no other assumptions, allowing applications in various settings. To improve the small sample performance, a bootstrap technique is proposed and theoretically justified. Based on this, we also present a procedure to simultaneously test the hypotheses of equal correlation and equal covariance matrices. The performance of all new test statistics is compared with existing procedures through extensive simulations.
We analyze the convergence properties of Fermat distances, a family of density-driven metrics defined on Riemannian manifolds with an associated probability measure. Fermat distances may be defined either on discrete samples from the underlying measure, in which case they are random, or in the continuum setting, in which they are induced by geodesics under a density-distorted Riemannian metric. We prove that discrete, sample-based Fermat distances converge to their continuum analogues in small neighborhoods with a precise rate that depends on the intrinsic dimensionality of the data and the parameter governing the extent of density weighting in Fermat distances. This is done by leveraging novel geometric and statistical arguments in percolation theory that allow for non-uniform densities and curved domains. Our results are then used to prove that discrete graph Laplacians based on discrete, sample-driven Fermat distances converge to corresponding continuum operators. In particular, we show the discrete eigenvalues and eigenvectors converge to their continuum analogues at a dimension-dependent rate, which allows us to interpret the efficacy of discrete spectral clustering using Fermat distances in terms of the resulting continuum limit. The perspective afforded by our discrete-to-continuum Fermat distance analysis leads to new clustering algorithms for data and related insights into efficient computations associated to density-driven spectral clustering. Our theoretical analysis is supported with numerical simulations and experiments on synthetic and real image data.
Sparse matrix operations involve a large number of zero operands which makes most of the operations redundant. The amount of redundancy magnifies when a matrix operation repeatedly executes on sparse data. Optimizing matrix operations for sparsity involves either reorganization of data or reorganization of computations, performed either at compile-time or run-time. Although compile-time techniques avert from introducing run-time overhead, their application either is limited to simple sparse matrix operations generating dense output and handling immutable sparse matrices or requires manual intervention to customize the technique to different matrix operations. We contribute a compile time technique called SpComp that optimizes a sparse matrix operation by automatically customizing its computations to the positions of non-zero values of the data. Our approach neither incurs any run-time overhead nor requires any manual intervention. It is also applicable to complex matrix operations generating sparse output and handling mutable sparse matrices. We introduce a data-flow analysis, named Essential Indices Analysis, that statically collects the symbolic information about the computations and helps the code generator to reorganize the computations. The generated code includes piecewise-regular loops, free from indirect references and amenable to further optimization. We see a substantial performance gain by SpComp-generated SpMSpV code when compared against the state-of-the-art TACO compiler and piecewise-regular code generator. On average, we achieve 79% performance gain against TACO and 83% performance gain against the piecewise-regular code generator. When compared against the CHOLMOD library, SpComp generated sparse Cholesky decomposition code showcases 65% performance gain on average.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191