In this work we present a deep convolutional neural network using 3D convolutions for Gait Recognition in multiple views capturing spatio-temporal features. A special input format, consisting of the gray-scale image and optical flow enhance color invariance. The approach is evaluated on three different datasets, including variances clothing, walking speeds view angle. contrast to most state-of-the-art systems used able generalize gait features across large angle changes. results show...

Summary A novel formulation of approximate truncated balanced realization (TBR) is introduced to unify three approaches: two iterative methods for solving the underlying Lyapunov equations - alternating directions implicit (ADI) iteration and rational Krylov subspace method (RKSM) a two-step procedure that performs Krylov-based projection subsequently direct TBR. The framework allows compare with respect solvability, fidelity, numerical effort, stability preservation global error bounds,...

We present a new approach to the problem of finding suitable expansion points in Krylov subspace methods for model reduction LTI systems. Using factorized formulation resulting error model, we can efficiently apply greedy algorithm and perform multiple steps instead looking all shifts at once. An expedient globally convergent optimization delivers locally ℋ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> -optimal two-dimensional ROMs...

Two approaches for approximating the solution of large-scale Lyapunov equations are considered: alternating direction implicit (ADI) iteration and projective methods by Krylov subspaces. We show that they linked in way ADI can always be identified a Petrov–Galerkin projection with rational block Therefore, unique Krylov-projected dynamical system associated iteration, which is proven to an H2 pseudo-optimal approximation. This includes generalisation previous results on pseudo-optimality...

In this article, a method to preserve stability in parametric model reduction by matrix interpolation is presented. Based on the measure approach, sufficient conditions original system matrices are derived. Once they fulfilled, of each reduced models guaranteed as well that resulting from interpolation. addition, it shown these met port-Hamiltonian systems and relevant set second-order obtained finite element method. The new approach illustrated two numerical examples.

Abstract This article introduces a novel way of transforming linear second order systems to first state space models in strictly dissipative realization. property extends the potential for analysis large-scale many ways: it allows application efficient methods solution Lyapunov equations and guaranteed preservation asymptotic stability during model reduction. The transformation is easy implement applicable very high due its negligible computational effort.

New sufficient conditions for ℌ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> pseudo-optimality in Krylov-based model reduction of linear dynamical systems are presented. The easy to evaluate and permit first applications: a new algorithm generate pseudo-optimal reduced models with respect the projecting subspace procedure superior local optima iterative methods. Numerical examples illustrate contributions.

We present rigorous bounds on the ℌ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> and xmlns:xlink="http://www.w3.org/1999/xlink">∞</sub> norm of error resulting from model order reduction second systems by KRYLOV subspace methods. To this end, we use a strictly dissipative state space realization perform factorization system. The derived expressions are easy to compute can therefore be applied models very high order, as is demonstrated...

In this paper a new formulation of the residual large-scale Lyapunov equations is presented, which results from approximate solution using projections byKrylov subspaces. The based on low-rank factors, allows an efficient numerical treatment even for equations. It shown, how matrix 2-norm can be computed with low effort. are presented most general case, means that generalized considered and oblique utilized approximately solving equation. With regard, also presents generalizations to...

Abstract In this paper, parametric model order reduction of linear time-invariant systems by matrix interpolation is adapted to large-scale in port-Hamiltonian form. A new weighted locally reduced models introduced preserve the structure, which guarantees passivity and stability interpolated system. The performance method demonstrated technical examples.

A method to preserve stability in parametric model order reduction by matrix interpolation for the whole parameter range is proposed high-order linear time-invariant systems. In first step, system matrices of high-dimensional parameter-dependent are computed a discrete set vectors. The local systems reduced projection-based method. Secondly, models made contractive solving low-dimensional Lyapunov equations. Thirdly, they transformed into consistent generalized coordinates accurate results....

Abstract This contribution summarizes a novel approach to error‐controlled model order reduction of second systems, i. e. new way find reduced specified approximation quality. To this end, starting from particular state space realization the model, one performs multiple steps and increments until (computationally affordable) rigorous global error bounds fall below prescribed limit. (© 2014 Wiley‐VCH Verlag GmbH &amp; Co. KGaA, Weinheim)