New geometric framework for positive semidefinite matrices of fixed rank.
problem Statistical analysis of positive semidefinite matrices of fixed rank.
method Introducing a manifold S(n,p)∗ with Riemannian geometry and Lie group structure. result Analytical closed forms for geodesics and Fréchet means.
New method learns multiple tasks efficiently by relaxing positive semidefinite constraint.
problem Efficiently learn multiple tasks with shared structure.
method Relax the positive semidefinite constraint on output kernel, solve unconstrained dual problem.
result Efficiently solve multi-task learning problems without eigendecomposition.
Proves stronger curvature condition for known nonnegative curvature manifolds.
problem Curvature conditions for manifolds with nonnegative sectional curvature.
method Modifies curvature operator with a 4-form to achieve positive-semidefiniteness.
result All known nonnegative curvature manifolds satisfy a stronger condition.
Paper tackles multi-label learning by improving SVR for positive semidefinite metrics.
problem Learning positive semidefinite metrics for multi-label and label distribution learning.
method Proposes two methods to overcome SVR's limitation in learning positive semidefinite metrics.
result Demonstrates new methods achieve favorable performance in multi-label and label distribution learning.
This article provides the mathematical foundation for stochastically continuous affine processes on the cone of positive semidefinite symmetric matrices. This analysis has been motivated by a large and growing use of matrix-valued affine processes in finance, including multi-asset option pricing with stochastic volatil…
Study finds polynomial convergence rate for Farey sequences linked to Riemann hypothesis.
problem Understanding convergence rates of maximum mean discrepancies for Farey sequences.
method Identifying positive-semidefinite kernels and their polynomial convergence rates.
result Polynomial convergence rate of maximum mean discrepancies of Farey sequences is equivalent to the Riemann hypothesis.
New inequalities for matrix supermartingales converge under various conditions.
problem Convergence and maximal inequalities of supermartingales in positive semidefinite matrices.
method Developed new concentration inequalities for matrix supermartingales.
result New inequalities for matrix supermartingales under different tail conditions.
New algorithm approximates large psd matrices from sketches.
problem Large-scale positive-semidefinite matrices from streaming data.
method Combines Nystrom approximation with rank truncation.
result Achieves prescribed relative error in Schatten 1-norm.
Matrix completion algorithms work well with random initialization.
problem Matrix completion with positive semidefinite constraints.
method Proved the absence of spurious local minima for non-convex optimization.
result Non-convex optimization algorithms can find global minima with arbitrary initialization.
New PSDMF algorithms derived from PR and ARM methods.
problem Positive semidefinite matrix factorization (PSDMF) challenges.
method Design PSDMF algorithms based on phase retrieval (PR) and affine rank minimization (ARM) methods.
result New PSDMF algorithms inherit numerical properties from PR and ARM methods.
Denise learns a function to quickly decompose covariance matrices robustly.
problem Robustly decomposing covariance matrices for feature extraction.
method Deep learning for symmetric positive semidefinite matrices.
result Denise achieves state-of-the-art performance in decomposition quality and speed.
Study nonconvex matrix completion for low-rank approximation without rank assumptions.
problem Low-rank approximation of positive semidefinite matrices from partial entries.
method Nonconvex optimization, local-minimum analysis, no spurious local minima.
result Improved sampling rate for nonconvex matrix completion with no spurious local minima.
A fast method estimates correlations in hybrid systems using observable market data.
problem Estimating instantaneous correlations in hybrid systems from observable data.
method Empirical correlations between observable market quantities are used to estimate state variables' correlations. Linear systems are involved, and the matrix is converted to positive semidefinite if necessary.
result The estimates are reasonably accurate, especially with more than 1,000 data points.
We consider a short rate model, driven by a stochastic process on the cone of positive semidefinite matrices. We derive sufficient conditions ensuring that the model replicates normal, inverse or humped yield curves.
The paper introduces new processes for modeling multivariate volatility.
problem Developing new stochastic processes for multivariate volatility modeling.
method Introducing Volterra Wishart and Volterra pure jump processes with fractional kernels.
result Affine covariance processes for multivariate volatility modeling.
We put forward a complete theory on moment explosion for fairly general state-spaces. This includes a characterization of the validity of the affine transform formula in terms of minimal solutions of a system of generalized Riccati differential equations. Also, we characterize the class of positive semidefinite process…
An algorithm for computing positive semidefinite factorizations of matrices.
problem Computing positive semidefinite factorizations of matrices.
method Non-commutative extension of Lee-Seung's algorithm (Matrix Multiplicative Update, MMU).
result The MMU algorithm ensures PSD updates and achieves critical points.
Over the past few years, trace regression models have received considerable attention in the context of matrix completion, quantum state tomography, and compressed sensing. Estimation of the underlying matrix from regularization-based approaches promoting low-rankedness, notably nuclear norm regularization, have enjoye…
Proves Gerber statistic is always non-negative.
problem Verifying the positive semi-definiteness of Gerber statistic.
method Analytical proof of both forms of Gerber statistic.
result Gerber statistic is positive semi-definite.
Optimal dictionaries minimize the average squared error in representing random vectors.
problem Finding optimal dictionaries for minimizing ℓ2-norm of coefficients in random vector representations. method Using rank-1 decompositions of symmetric positive semidefinite matrices, explicit descriptions and polynomial-time algorithms for ℓ2-optimal dictionaries are provided. result Explicit descriptions and polynomial-time algorithms for ℓ2-optimal dictionaries are provided. A streaming algorithm estimates quadratic covariation from financial data efficiently.
problem Estimating quadratic covariation from ultra-high-frequency financial data with limited memory.
method Formulated multi-scale, realized kernel, pre-averaging, and modulated realized covariance estimators with fixed bandwidth.
result Fixed bandwidth estimators require higher bandwidth for positive semidefiniteness.
A new matrix concentration inequality for random products of matrices.
problem Understanding the behavior of random matrix products under bounded independent positive semidefinite matrices.
method Developed a non-asymptotic concentration inequality for the product of matrices.
result The inequality provides a bound on the deviation of the matrix product from its expected value.
Most existing word embedding methods can be categorized into Neural Embedding Models and Matrix Factorization (MF)-based methods. However some models are opaque to probabilistic interpretation, and MF-based methods, typically solved using Singular Value Decomposition (SVD), may incur loss of corpus information. In addi…
Introduces a new model for mapping matrices to matrices, subsuming linear regression.
problem Learning matrix-to-matrix mappings from data.
method Partial trace regression model, leveraging quantum information theory.
result Relevance demonstrated in matrix-to-matrix regression and positive semidefinite matrix completion.
New kernel improves graph classification accuracy.
problem Improving graph classification accuracy.
method Developed an optimal assignment kernel for graphs.
result Improved graph classification accuracy on benchmark data.
Global stability bounds for matrix frames in phase retrieval problems.
problem Phase retrieval for matrix frames in various applications.
method Computable global stability bounds for the quasi-linear analysis map β, using Whitney stratification of positive semidefinite matrices of low rank.
result Novel conditions for a frame to be generalized phase retrievable.
Improved covariance matrix estimation for portfolio optimization with guaranteed PSD and controlled conditioning.
problem Guaranteeing positive semidefinite ness and controlling spectral conditioning in IQ estimators.
method Introducing squeezing identity and atomic-IQ parameterization to construct structured channel matrices with PSD guarantees and analytic eigen floor for conditioning control.
result Atomic-IQ improves Sharpe ratios and delivers a more stable risk profile compared to standard estimators.
We glue two manifolds which have curvature operators at least k (in the sense of eigenvalues) along their common boundary. We show that if the sum of the second fundamental forms of the boundary is positive semidefinite, then the curvature operator of the resulting manifold is at least k up to an arbitrarily small erro…
We propose a simple, scalable, and fast gradient descent algorithm to optimize a nonconvex objective for the rank minimization problem and a closely related family of semidefinite programs. With O(r3κ2nlogn) random measurements of a positive semidefinite n×n matrix of rank r and condition number κ…
We present a hybrid algorithm for optimizing a convex, smooth function over the cone of positive semidefinite matrices. Our algorithm converges to the global optimal solution and can be used to solve general large-scale semidefinite programs and hence can be readily applied to a variety of machine learning problems. We…
Paper analyzes convergence of distributed inference using BP in linear Gaussian models.
problem Distributed inference convergence in linear Gaussian models.
method Factor graphs, Gaussian belief propagation, local computation, message passing.
result Message information matrix converges to a unique positive definite limit matrix at a doubly exponential rate.
Trace norm regularization is a popular method of multitask learning. We give excess risk bounds with explicit dependence on the number of tasks, the number of examples per task and properties of the data distribution. The bounds are independent of the dimension of the input space, which may be infinite as in the case o…
Let G be a compact connected Lie group and H a closed subgroup of G. Suppose the homogeneous space G/H is effective and has dimension 3 or higher. Consider a G-invariant, symmetric, positive-semidefinite, nonzero (0,2)-tensor field T on G/H. Assume that H is a maximal connected Lie subgroup of G. We p…
Accelerated RPCholesky speeds up kernel matrix approximations.
problem Efficiently approximating large kernel matrices.
method Accelerated randomly pivoted Cholesky (RPCholesky) with block matrix computations and rejection sampling.
result Approximates kernel matrices up to 40 times faster.
Proposes a Riemannian optimization for policy improvement in MDPs.
problem Optimizing policy functions in Markov decision processes (MDPs).
method Riemannian proximal optimization algorithm with Gaussian mixture model (GMM).
result Guaranteed convergence and efficacy demonstrated in preliminary experiments.
Solves matrix completion for rectangular matrices using gradient descent.
problem Matrix completion for rectangular matrices with limited data.
method Lifts matrix to positive semidefinite, optimizes over semidefinite factor using gradient descent.
result Algorithm converges linearly to global optimum with high probability.
Improved guarantees for nonconvex matrix factorization with rank overparameterization.
problem Minimizing nonconvex objective over low-rank matrices.
method Overparameterized Burer--Monteiro approach, leveraging smoothness and strong convexity.
result Local optimization globally converges to global optimum under certain rank conditions.
In this paper we demonstrate that under general conditions there exists a metric in the conformal class of an arbitrary metric on a smooth, closed Riemannian manifold of dimension greater than four such that the Q-curvature of the metric is a constant. Existence of solutions is obtained through the combination of var…
The paper introduces algorithms for efficient low-rank matrix approximation.
problem Efficiently approximating large matrices while preserving their properties.
method Random linear images (sketches) of the matrix, with error bounds for quality control.
result Simple, accurate, numerically stable methods for low-rank approximation.
This paper addresses the problem of low-rank distance matrix completion. This problem amounts to recover the missing entries of a distance matrix when the dimension of the data embedding space is possibly unknown but small compared to the number of considered data points. The focus is on high-dimensional problems. We r…
This work deals with the simulation of Wishart processes and affine diffusions on positive semidefinite matrices. To do so, we focus on the splitting of the infinitesimal generator, in order to use composition techniques as Ninomiya and Victoir or Alfonsi. Doing so, we have found a remarkable splitting for Wishart proc…
Improved stability for matrix recovery from rank-one measurements.
problem Phase retrieval problem of recovering rank-one positive semidefinite matrices.
method Developed a smoothing Newton method based on Bures-Wasserstein gradient descent.
result Superlinear convergence with rigorous guarantees and stable implementation.
Bayesian optimization gains efficiency by leveraging symmetries through a modified max kernel.
problem Improving Bayesian optimization efficiency for functions with group symmetries.
method Developed a PSD projection of the max kernel to exploit symmetries without violating kernel properties.
result The modified max kernel achieves lower regret compared to existing invariant and non-invariant kernels.
The computation of the sparse principal component of a matrix is equivalent to the identification of its principal submatrix with the largest maximum eigenvalue. Finding this optimal submatrix is what renders the problem NP-hard. In this work, we prove that, if the matrix is positive semidefinite and its …
We are concerned with an approximation problem for a symmetric positive semidefinite matrix due to motivation from a class of nonlinear machine learning methods. We discuss an approximation approach that we call {matrix ridge approximation}. In particular, we define the matrix ridge approximation as an incomplete matri…
Recent research in off-the-grid compressed sensing (CS) has demonstrated that, under certain conditions, one can successfully recover a spectrally sparse signal from a few time-domain samples even though the dictionary is continuous. In particular, atomic norm minimization was proposed in \cite{tang2012csotg} to recove…
Semidefinite tests detect latent causal structures efficiently.
problem Testing causal relations in the presence of latent variables.
method Semidefinite programming to test the signature of latent structures in observable covariance matrices.
result Semidefinite tests are computationally efficient and can detect latent causal structures.
RPCholesky approximates kernel matrices with few evaluations.
problem Approximating kernel matrices efficiently.
method Randomly pivoted partial Cholesky factorization.
result RPCholesky provides nearly optimal low-rank approximations.