On numerical approximation schemes for expectation propagation

نویسنده

  • Alexis Roche
چکیده

Several numerical approximation strategies for the expectation-propagation algorithm are studied in the context of large-scale learning: the Laplace method, a faster variant of it, Gaussian quadrature, and a deterministic version of variational sampling (i.e., combining quadrature with variational approximation). Experiments in training linear binary classifiers show that the expectation-propagation algorithm converges best using variational sampling, while it also converges well using Laplacestyle methods with smooth factors but tends to be unstable with non-differentiable ones. Gaussian quadrature yields unstable behavior or convergence to a sub-optimal solution in most experiments.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Biorthogonal wavelet-based full-approximation schemes for the numerical solution of elasto-hydrodynamic lubrication problems

Biorthogonal wavelet-based full-approximation schemes are introduced in this paper for the numerical solution of elasto-hydrodynamic lubrication line and point contact problems. The proposed methods give higher accuracy in terms of better convergence with low computational time, which have been demonstrated through the illustrative problems.

متن کامل

Verification and Validation of Common Derivative Terms Approximation in Meshfree Numerical Scheme

In order to improve the approximation of spatial derivatives without meshes, a set of meshfree numerical schemes for derivative terms is developed, which is compatible with the coordinates of Cartesian, cylindrical, and spherical. Based on the comparisons between numerical and theoretical solutions, errors and convergences are assessed by a posteriori method, which shows that the approximations...

متن کامل

Nonstandard finite difference schemes for differential equations

In this paper, the reorganization of the denominator of the discrete derivative and nonlocal approximation of nonlinear terms are used in the design of nonstandard finite difference schemes (NSFDs). Numerical examples confirming then efficiency of schemes, for some differential equations are provided. In order to illustrate the accuracy of the new NSFDs, the numerical results are compared with ...

متن کامل

Approximate inference techniques with expectation constraints

This article discusses inference problems in probabilistic graphical models that often occur in a machine learning setting. In particular it presents a unified view of several recently proposed approximation schemes. Expectation consistent approximations and expectation propagation are both shown to be related to Bethe free energies with weak consistency constraints, i.e. free energies where lo...

متن کامل

Spectral analysis of finite-difference approximations of 1−d waves in non-uniform grids

Preserving the finite positive velocity of propagation of continuous solutions of wave equations is one of the key issues, when building numerical approximation schemes for control and inverse problems. And this is hard to achieve uniformly on all possible ranges of numerical solutions. In particular, high frequencies often generate spurious numerical solutions, behaving in a pathological manne...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:
  • CoRR

دوره abs/1611.04416  شماره 

صفحات  -

تاریخ انتشار 2016